An unmanned delivery system based on intelligent weighing

Abstract

The present application relates to the technical field of intelligent logistics and unmanned delivery, and particularly relates to an unmanned delivery system based on intelligent weighing, which comprises a multi-modal data acquisition module, a distributed weighing and dynamic optimization module, a loading execution module, a detection and verification module, and an electronic delivery order generation module; the system calculates the loading sequence and path of goods through a dynamic loading optimization model, and monitors the change of the center of gravity and the load distribution of the vehicle by using an inertial sensor and a piezoelectric sensor to realize real-time adjustment and optimization; through the combination of the electronic delivery order and the digital twin entity, the system supports full-process state tracking and abnormal early warning to ensure the safety and management efficiency of transportation; the present application significantly improves the intelligence and efficiency of logistics operation, and provides comprehensive technical support for modern logistics and factory delivery.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics and unmanned shipping technology, and in particular to an unmanned shipping system based on intelligent weighing. Background Technology

[0002] Unmanned shipping systems are crucial for modern logistics and factory cargo shipping. The shipping process is a key link in the goods circulation process, and its efficiency and accuracy directly impact a company's operational efficiency and logistics costs. In traditional shipping methods, manual operation is prone to errors, and the inefficiency of vehicles entering and leaving the factory area leads to congestion and management chaos. Existing technology (Chinese Invention Patent, Publication No.: CN113108884A, Title: An AI Unmanned Weighing and Shipping Method) mainly achieves this through the following technical means: using RFID cards and remote card readers to record and identify vehicle information; using a weighbridge in a fixed weighing area to weigh the initial weight and the weight after loading of the vehicle; and storing and verifying transportation data through a cloud database. However, this technology has the following drawbacks:

[0003] Existing solutions rely solely on weight differences to determine vehicle loading status, failing to monitor the vehicle's center of gravity, which may lead to vehicle instability and safety hazards during loading. Furthermore, unforeseen circumstances during loading (such as incorrect cargo placement or weight deviation) cannot be dynamically adjusted in real time to change the loading sequence or route, resulting in low system adaptability. The synchronization of cargo information recorded by RFID cards with the cloud database lacks real-time monitoring and feedback mechanisms, making it unable to respond to dynamic changes in the logistics process. Finally, the loading process does not consider cargo dimensions and the vehicle's spatial model, potentially leading to wasted space or improper loading. Summary of the Invention

[0004] To address the numerous problems existing in the aforementioned technologies, this invention provides an unmanned shipping system based on intelligent weighing. This invention utilizes a technical framework of multimodal data acquisition, dynamic optimization, real-time monitoring, and end-to-end tracking. It leverages distributed weighing and dynamic optimization models, combining reinforcement learning and cellular automata rules to optimize the loading sequence and path of goods. Simultaneously, dynamic center of gravity monitoring and deviation verification ensure loading safety, while digital twin entity synchronization technology enables end-to-end status tracking and management. This invention significantly improves logistics efficiency, load balancing, and transportation safety.

[0005] An unmanned shipping system based on intelligent weighing includes:

[0006] The multimodal data processing module is used to collect multimodal data, including vehicle feature data, driver feature data, and cargo feature data; to perform quantum encoding on the multimodal data; to map the multimodal data to a high-dimensional space using the kernel function method; to extract the core feature correlations between the data through high-dimensional correlation analysis; to generate correlated feature data; and to generate digital twin entity data by combining task information.

[0007] The distributed weighing and optimization module is used to calculate the weight distribution model through vehicle contact pressure data, and combine vehicle specification information and cargo characteristic data to generate a dynamic loading optimization model using reinforcement learning optimization algorithms.

[0008] The loading execution module includes a loading robot, which generates loading task data based on dynamic loading optimization model data and performs cargo loading operations; it monitors changes in the vehicle's center of gravity through inertial sensors and captures load changes during cargo loading through piezoelectric sensors, and adjusts the loading sequence and path in real time in conjunction with a dynamic interactive prediction algorithm to optimize the loading process; the dynamic center of gravity distribution data generated during the loading process and the cargo placement status data together constitute the loading status data.

[0009] The detection and verification module is used to collect vehicle detection data after loading is completed, generate verification results based on the comparison between the detection data and the loading status data, and generate an electronic shipping document when the verification is successful.

[0010] Preferably, the spectral scanning device in the multimodal data processing module emits and receives spectral data reflected from the vehicle surface through a multi-band light source, and generates vehicle spectral feature data by combining it with a preset spectral feature matching algorithm; the facial recognition camera classifies and extracts the driver's facial features through a convolutional neural network algorithm to generate driver facial feature data; and the three-dimensional laser scanner obtains three-dimensional dimension data of the cargo surface through laser beam ranging technology to generate cargo dimension feature data.

[0011] Preferably, the chemical sensor in the multimodal data processing module detects the molecular composition of the cargo material using gas-phase molecular analysis, and generates material property description data of the cargo by combining it with the material feature model matching algorithm stored in the database. This data is then input into the multimodal data processing module as part of the cargo feature data.

[0012] Preferably, the pressure sensor array in the distributed weighing and optimization module is installed in a grid-like arrangement along the vehicle's driving path. By collecting pressure data at the vehicle's tire contact points in real time and combining it with the vehicle's axle load distribution model, the weight distribution data of the vehicle is generated.

[0013] Preferably, the distributed weighing and optimization module utilizes a deep Q-network structure in the reinforcement learning optimization algorithm, inputs vehicle weight distribution data, cargo three-dimensional size feature data and vehicle specification information, and outputs a dynamic loading optimization model for the vehicle. The training objectives of the reinforcement learning optimization algorithm include minimizing loading balance, center of gravity stability and loading time efficiency.

[0014] Preferably, the dynamic loading optimization model is simulated dynamically through cellular automata rules. The cell state includes the current cargo position, the vehicle center of gravity position, and the cargo priority. The cell state change rules include the impact of the newly added cargo position on the center of gravity of neighboring cells. The simulation results include loading task data and cargo priority ranking.

[0015] Preferably, the loading robot in the loading execution module has a six-axis robotic arm structure, and combined with a path planning algorithm, performs precise grasping and placement operations of goods according to the loading task data; the loading robot monitors the changes in the vehicle's center of gravity in real time through inertial sensors, and optimizes the loading path by combining a dynamic interactive prediction algorithm, wherein the dynamic interactive prediction algorithm is based on a recurrent neural network to predict the impact of loading sequence adjustments.

[0016] Preferably, the dynamic interactive prediction algorithm combines cargo size feature data and vehicle loading space model data to recalculate loading sequence and path optimization parameters after each cargo placement operation, generating real-time adjusted loading task data.

[0017] Preferably, the dynamic detection device in the detection and verification module collects the actual load distribution data of the vehicle through a pressure sensor, and generates the actual center of gravity data and vibration mode data of the vehicle by combining it with an inertial sensor and a vibration sensor. The comparison algorithm calculates the deviation range based on the vehicle design load model and loading state data, and generates verification results based on the deviation range.

[0018] Preferably, the electronic waybill includes vehicle information, cargo information, shipping time, verification results, and deviation range evaluation indicators. The electronic waybill is synchronized to the transportation management system through a digital twin entity for status tracking of subsequent logistics operations.

[0019] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0020] This invention achieves load balance optimization through dynamic center of gravity monitoring technology: by using inertial sensors and dynamic center of gravity calculation methods, the changes in the vehicle's center of gravity are monitored in real time to ensure the stability and safety of the vehicle during loading.

[0021] This invention achieves real-time adjustment of the loading process through a dynamic interactive prediction algorithm: by combining cargo size feature data and loading space model data, the loading sequence and path parameters are dynamically optimized, significantly improving loading efficiency and adaptability.

[0022] This invention enables full-process status tracking of logistics through digital twin entities: by synchronizing electronic shipping documents with digital twin entities, dynamic monitoring and early warning of anomalies during transportation are achieved.

[0023] This invention maximizes the utilization of cargo space through spatial matching optimization technology: based on the precise matching of cargo characteristics and loading space, it avoids the waste of loading space. Attached Figure Description

[0024] Figure 1 This is a structural block diagram of the system of the present invention;

[0025] Figure 2 This is a schematic diagram of the workflow of the multimodal data processing module in this invention;

[0026] Figure 3 This is a schematic diagram of the dynamic optimization workflow of the loading execution module in this invention;

[0027] Figure 4 This is a flowchart illustrating the detection and verification module in this invention. Detailed Implementation

[0028] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0029] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0031] like Figure 1As shown, an unmanned shipping system based on intelligent weighing includes:

[0032] like Figure 2 As shown, the multimodal data processing module is used to collect multimodal data, including vehicle feature data, driver feature data, and cargo feature data; to perform quantum encoding on the multimodal data; to map the multimodal data to a high-dimensional space using the kernel function method; to extract the core feature associations between the data through high-dimensional correlation analysis; to generate associated feature data; and to generate digital twin entity data by combining task information.

[0033] The collection of multimodal data includes three categories: vehicle feature data, driver feature data, and cargo feature data, each corresponding to different collection devices and technical approaches:

[0034] Vehicle feature data is obtained by using a spectral scanning device to emit and receive reflectance spectral data from a multi-band light source onto the vehicle surface. A spectral feature matching algorithm (e.g., a correlation analysis-based algorithm) is then used to classify the spectral data, generating the vehicle's spectral feature data.

[0035] Driver feature data is collected by facial recognition cameras, and classified using a convolutional neural network (CNN) model to extract driver facial feature data. At the same time, a voice sensor is used to collect driver voice signals and extract their frequency features as supplementary data.

[0036] Cargo feature data is generated by using a 3D laser scanner to capture the 3D shape and dimensions of the cargo; chemical sensors detect the volatile gas composition on the surface of the cargo to generate chemical property description data.

[0037] The acquired multimodal data is quantum-encoded, and a kernel function method is used to map the nonlinear data to a high-dimensional feature space, capturing the implicit correlations in the data. The kernel function mapping method is implemented using a Gaussian kernel function, which maps the input data... Transformed into a mapping vector in the feature space The calculation formula is as follows:

[0038]

[0039] in, For input data points; Represents Euclidean distance; The bandwidth parameter of the kernel function controls the amplitude decay of the Gaussian function. This mapping transforms the nonlinear characteristics of multimodal data into a linear distribution in a high-dimensional space, facilitating subsequent analysis.

[0040] In high-dimensional space, singular value decomposition (SVD) is used to extract core correlation features among multimodal data. The core steps of this method include:

[0041] Constructing a multimodal data matrix Each column in the matrix represents a type of feature data. Perform singular value decomposition: ,in: and These are the left singular matrix and the right singular matrix, respectively; Given a diagonal matrix, the values ​​on the diagonal are singular values. Select the largest of the following... Each singular value and its corresponding singular vector are used to generate associated feature data, which is used to represent the core association between data.

[0042] After extracting the relevant feature data, it is combined with task information (including transportation task number, cargo list, planned loading information, etc.) to generate digital twin entity data. The digital twin entity is a dynamically updated virtual model that reflects the status of vehicles, cargo, and tasks in real time, providing global data support for subsequent modules.

[0043] Example Application: A logistics company uses this system for unmanned shipments. When a truck enters the factory area: a spectral scanning device captures the vehicle's external spectral characteristics (such as paint type and damage). A facial recognition camera confirms the driver's identity, and a voice sensor verifies the driver's voice characteristics. The cargo's dimensions (length × width × height) are measured as 2.5m × 1.8m × 1.2m by a 3D laser scanner, and a chemical sensor detects the cargo's surface composition as polypropylene. The collected multimodal data is quantum encoded, and a feature vector is generated using a Gaussian kernel function mapping. High-dimensional correlation analysis: singular value decomposition is used to extract correlation features between multimodal data. The correlation results show that the vehicle's load capacity matches the cargo size with a degree of 0.95 (range 0-1, 1 being a perfect match). The driver's operating habits have a weight of 0.7 on the impact of cargo vibration. The above results are integrated into a digital twin entity, including the correlation information of the vehicle, driver, and cargo, and synchronized to the optimization module in real time.

[0044] Preferably, the spectral scanning device in the multimodal data processing module emits and receives spectral data reflected from the vehicle surface through a multi-band light source, and generates vehicle spectral feature data by combining it with a preset spectral feature matching algorithm; the facial recognition camera classifies and extracts the driver's facial features through a convolutional neural network algorithm to generate driver facial feature data; and the three-dimensional laser scanner obtains three-dimensional dimension data of the cargo surface through laser beam ranging technology to generate cargo dimension feature data.

[0045] In this invention, a multimodal data processing module is used to accurately collect and digitize the characteristics of vehicles, drivers, and goods, supporting subsequent data processing and optimization in the intelligent weighing system. This module consists of a spectral scanning device, a facial recognition camera, and a 3D laser scanner, which are respectively used for collecting feature data of vehicles, drivers, and goods.

[0046] The spectral scanning device works by emitting light beams onto the vehicle surface using a multi-band light source, while simultaneously acquiring reflectance spectral data through a receiving device. This reflectance spectral data reflects the optical properties of the vehicle surface materials, such as coating color, texture, and damage. To improve accuracy, the device incorporates a spectral feature matching algorithm to analyze and classify the acquired spectral data.

[0047] Spectral feature matching algorithm, based on matching score calculation:

[0048]

[0049] in, This represents the matching score, used to determine spectral similarity; This represents the actual spectral reflectance collected; Represents the reference reflectance of the preset spectral model; This indicates the number of spectral channels. Ultimately, the device outputs vehicle spectral characteristic data, providing input for subsequent load distribution analysis and vehicle condition monitoring.

[0050] Facial recognition cameras use convolutional neural network (CNN) algorithms to classify and extract facial features of drivers. The specific implementation process includes capturing images of the driver's face, extracting multi-level local features (such as texture information of the eyes, nose, and mouth) through convolutional layers, and outputting feature vectors through fully connected layers to generate driver facial feature data.

[0051] Using the grayscale values ​​of a facial image as input, feature maps are calculated through layer-by-layer convolution operations, ultimately generating a driver identity feature vector in the classification layer. For example, the driver's feature extraction output vector is:

[0052]

[0053] in, The feature vector representing the driver; This represents numerical values ​​for different feature dimensions. The output data is used to verify driver identity and record driving behavior characteristics.

[0054] A 3D laser scanner captures three-dimensional dimensional data of a cargo surface using laser beam ranging technology. The device measures the time difference between the laser beam's emission and return by rotating a laser emitter and receiver, and calculates spatial distance using triangulation principles. The laser ranging formula is:

[0055]

[0056] in, This indicates the measured distance; Represents the speed of light; This represents the time difference between laser emission and reception. Using the ranging results from continuous rotation, three-dimensional point cloud data of the cargo can be reconstructed, and cargo size feature data can be generated, accurately describing the cargo's length, width, height, and surface shape.

[0057] Precise analysis of the optical properties of vehicle surfaces is used to detect paint condition and surface damage, preventing vehicle identification errors during weighing and shipping. Convolutional neural networks are used to extract driver facial features, ensuring accurate and safe driver identification. Laser ranging technology generates three-dimensional dimensional feature data of cargo, providing necessary input for loading planning and load balancing.

[0058] In an example, in an unmanned delivery system, a truck enters the factory area for multimodal data collection. A spectral scanning device emits a multi-band light source, receives the reflected spectrum from the vehicle's surface, and analyzes it to determine that the vehicle's surface coating is metallic paint, with surface damage accounting for 2% of the total coating area. A facial recognition camera captures the driver's facial feature image, extracts feature vectors using a CNN algorithm, and compares them with the driver database in the system to confirm the driver's identity as "Zhang San". A 3D laser scanner acquires the 3D point cloud data of the cargo, and after analysis, generates the cargo's length, width, and height as 2.5m, 1.8m, and 1.2m, respectively. The cargo surface has no indentations and is a regular cuboid. Finally, the above data is aggregated and fed into the multimodal data processing module to generate multimodal feature data for the vehicle, driver, and cargo, which serves as data input for the digital twin entity and is used by the system's subsequent optimization modules.

[0059] Preferably, the chemical sensor in the multimodal data processing module detects the molecular composition of the cargo material using gas-phase molecular analysis, and generates material property description data of the cargo by combining it with the material feature model matching algorithm stored in the database. This data is then input into the multimodal data processing module as part of the cargo feature data.

[0060] The chemical sensor detects volatile gas molecules released from the surface of cargo using gas-phase molecular analysis. Specifically, the sensor draws volatile gas molecules into a detection chamber, converts them into measurable ionic signals using an ionization device, and then analyzes the mass-to-charge ratio (m / z) of the ions using a mass analyzer to generate characteristic data of the molecular composition.

[0061] Mass-to-charge ratio analysis: The detection formula is as follows:

[0062]

[0063] in, Indicates the mass-to-charge ratio; Indicates ion mass; This indicates the ionic charge. This method enables the rapid and accurate separation and identification of the main molecular components of cargo materials, such as polypropylene (PP) or polyethylene (PE) in plastics.

[0064] The detected molecular composition data needs to be further matched with material feature models in the system database to generate cargo material property description data. This algorithm selects the most suitable material type by calculating the similarity of molecular property vectors. The property vector matching formula is as follows:

[0065]

[0066] in, This indicates the similarity score; Let the first and second parts of the detection feature vector and the database feature vector be respectively the first and second parts of the detection feature vector and the database feature vector. One component; The dimension of the feature vector; This represents the cosine value of the angle between feature vectors. When the similarity score... When the system exceeds the matching threshold set by the system, the system will mark the material property description data as the matching material type and input the matching result as part of the cargo feature data into the multimodal data processing module.

[0067] The detection results from chemical sensors are integrated with data from other modules (such as a 3D laser scanner) to form complete cargo characteristic data. For example, combining the cargo's material properties with its 3D dimensions can provide more accurate input for subsequent loading optimization models.

[0068] In an example, during a logistics transportation task, a batch of goods needs to be weighed and loaded using an unmanned dispatch system. The goods contain various materials with complex compositions and similar appearances. To ensure the accuracy of the loading plan, the system utilizes a chemical sensor module. The chemical sensor collects volatile gas molecules from the surface of the goods and generates ion signals through an ionization device. The main ion characteristics detected by the quality analyzer are as follows:

[0069] mass-to-charge ratio Corresponding to isobutylene;

[0070] mass-to-charge ratio Corresponding polypropylene molecular fragment.

[0071] Based on comprehensive analysis of molecular signals, the system confirmed that the main material of the cargo is polypropylene (PP).

[0072] The detection results are compared with the database model using a feature vector matching algorithm to obtain a similarity score. The value is higher than the system's set matching threshold of 0.9. The system confirms that the cargo material is polypropylene (PP) and generates "polypropylene material" characteristic description data. The generated material characteristic description data, along with the cargo dimensions (length 2.5m, width 1.8m, height 1.2m) measured by the 3D laser scanner, are input into the multimodal data processing module to provide complete cargo characteristic data for the loading optimization module.

[0073] The distributed weighing and optimization module is used to calculate the weight distribution model through vehicle contact pressure data, and combine vehicle specification information and cargo characteristic data to generate a dynamic loading optimization model using reinforcement learning optimization algorithms.

[0074] As the vehicle enters the weighing area, an array of pressure sensors deployed on the ground collects real-time pressure data at the tire contact points. The pressure sensors are arranged in a grid, with each node corresponding to a sampling point, enabling the capture of detailed features of the vehicle's weight distribution. Weight distribution model calculation: Based on a physical model of the vehicle's axle load distribution, the collected pressure data is transformed into a weight distribution model.

[0075]

[0076] in, Indicates the total weight of the vehicle; Indicates the first Pressure values ​​at each sensor node; Indicates the first The force-bearing area corresponding to each node; This represents the total number of sensor nodes. Ultimately, the weight distribution model describes the load distribution across the vehicle's axles, providing fundamental data for load balance analysis.

[0077] The system integrates cargo characteristic data (such as dimensions and material properties) obtained through the multimodal data processing module with vehicle specification information (such as maximum load, wheelbase, and cargo compartment dimensions). It then overlays and simulates the cargo size and weight data to analyze the impact of different cargo placement schemes within the cargo compartment on weight distribution, generating a preliminary load distribution strategy.

[0078] With the support of the aforementioned basic data, the system employs reinforcement learning algorithms to optimize the loading strategy and generate a dynamic loading optimization model. The main logic of reinforcement learning is to find the optimal loading sequence and position through repeated trials (iterations) to maximize the optimization of vehicle load balance and loading efficiency.

[0079] After optimization, the system generates a dynamic loading optimization model, which includes key information such as the order of cargo placement, specific location coordinates, and expected weight distribution. This model serves as input to the loading execution module, guiding the loading robot to complete subsequent operations.

[0080] Preferably, the pressure sensor array in the distributed weighing and optimization module is installed in a grid-like arrangement along the vehicle's driving path. By collecting pressure data at the vehicle's tire contact points in real time and combining it with the vehicle's axle load distribution model, the weight distribution data of the vehicle is generated.

[0081] The pressure sensor array is installed in a grid pattern along the vehicle's travel path to ensure that the sensors cover all tire contact areas. Each sensor node measures the pressure value of the vehicle's tires at the contact point, acquiring the distribution of the vehicle's load on the sensor plane. The sensor nodes are arranged at fixed intervals, with each node representing the pressure value of a grid cell. As the vehicle travels through the sensor array, each node records real-time pressure data. ,in The nodes are assigned numbers. This data is uploaded to the processing module in real time via the data acquisition unit.

[0082] By combining the collected pressure data with the vehicle's geometric characteristics (such as tire position and axle spacing), the system uses a vehicle axle weight distribution model to calculate the weight distribution of the vehicle on each axle. The contact pressure value of each tire has a linear relationship with the total load on its corresponding axle, as shown in the following formula:

[0083]

[0084] in, Indicates the first The total weight of each axle; Indicates the first Pressure values ​​at each sensor node; Represents nodes The corresponding effective contact area; This indicates the number of sensor nodes belonging to that axle. By aggregating the weight data of all axles, the system generates an overall weight distribution model of the vehicle.

[0085] Weight distribution data includes axle load data and the vehicle's center of gravity position. The system generates this data through the following steps: axle load data, calculating the load proportion of each axle to determine whether the actual load on each axle exceeds the vehicle's design specifications; center of gravity position calculation, combining load data and axle spacing to calculate the vehicle's center of gravity position, with the output result used to assess the balance of load distribution.

[0086] The sensor array can collect vehicle contact pressure data in real time, ensuring the system can quickly reflect the vehicle's weight distribution. Through gridded layout and comprehensive analysis of multi-node pressure data, the weight distribution model can accurately describe the load on each axle and the vehicle's center of gravity. By dynamically monitoring the vehicle's weight distribution, overloading or uneven load distribution can be identified in a timely manner, ensuring vehicle safety during subsequent transportation. Weight distribution data provides crucial input to the system's loading optimization module, helping to generate efficient loading plans.

[0087] In an example, a logistics company uses a distributed weighing and optimization module to weigh and analyze the load of a truck in its unmanned dispatching system.

[0088] As the vehicle travels over the pressure sensor array, the sensors collect the pressure value at each tire contact point. Assume the pressure data for the front tires is... and The pressure data for the rear wheels is and The sensor array automatically transmits the pressure values ​​to the data processing module.

[0089] The system combines the effective contact area of ​​the sensor nodes (assuming it is ). Calculate the shaft load:

[0090]

[0091]

[0092] Calculate the total weight of the vehicle. .

[0093] The vehicle's center of gravity position is calculated using axle spacing and load data to confirm whether the vehicle's load distribution is uniform.

[0094] The weight distribution data is generated, including: the front axle load is 500 kg, accounting for 32.89% of the total load; the rear axle load is 1020 kg, accounting for 67.11% of the total load; and the center of gravity is located 65% of the rear of the vehicle's wheelbase.

[0095] Preferably, the distributed weighing and optimization module utilizes a deep Q-network structure in the reinforcement learning optimization algorithm, inputs vehicle weight distribution data, cargo three-dimensional size feature data and vehicle specification information, and outputs a dynamic loading optimization model for the vehicle. The training objectives of the reinforcement learning optimization algorithm include minimizing loading balance, center of gravity stability and loading time efficiency.

[0096] Reinforcement learning is an adaptive optimization method based on a trial-and-error mechanism. Deep Q-networks, by combining traditional Q-learning with deep neural networks, enhance optimization capabilities in high-dimensional state spaces. In this invention:

[0097] The state space includes the vehicle's current weight distribution, the cargo's pending loading status (position, size, weight, etc.), and the available space distribution in the cargo compartment. Each loading operation includes selecting the specific location and order for cargo placement. Based on three objectives—loading balance, center of gravity stability, and loading time efficiency—the reward function is specifically defined as follows:

[0098]

[0099] in, Indicates the state Take action below The reward value obtained; This indicates the load balance score, which measures the degree of balance between the front and rear axle loads of the vehicle. This represents the center of gravity stability score, which measures the stability of the vehicle's load center of gravity shift. This indicates the loading time consumed and is used to penalize inefficient operations. , , These are represented as weight coefficients for each objective, and are dynamically adjusted according to actual needs.

[0100] The Deep Q-Network processes input data using a Convolutional Neural Network (CNN) structure: vehicle weight distribution data, input as a two-dimensional matrix representing the current load state of each area within the cargo compartment; three-dimensional dimensional feature data of the cargo, provided by a multimodal data processing module, which is then input into the CNN and transformed into a standardized dimensional feature vector; and vehicle specification information, including maximum cargo compartment load and size limitations, which are incorporated into the optimization process as constraints.

[0101] In each round of decision-making, the deep Q-network calculates the Q-value of all possible actions based on the current state and selects the action with the highest Q-value as the optimal decision. The dynamic loading optimization model includes the following core components: the placement location of each item (…). Coordinates); loading sequence (dynamically adjusted based on cargo characteristics and spatial layout); expected load distribution and center of gravity location.

[0102] Leveraging deep learning technology, the system can rapidly generate efficient loading plans within complex, multi-dimensional state spaces. By optimizing loading positions, it significantly reduces front and rear axle load deviations, ensuring transportation safety. Dynamically adjusting the order and position of cargo placement prevents vehicle rollovers or loss of control due to center of gravity shifts. Based on reward function optimization for loading time efficiency, the system effectively reduces loading time and improves overall logistics efficiency.

[0103] In an example, a logistics company needs to load a batch of goods into a truck. The goods have the following characteristics:

[0104] Goods A: Weight 1,500 kg, dimensions 2m × 1.5m × 1.2m;

[0105] Goods B: Weight 800kg, dimensions 1.8m × 1.2m × 1.0m;

[0106] Goods C: Weight 1,200kg, dimensions 1.6m×1.4m×1.1m.

[0107] Input data: Vehicle weight distribution data, with a front axle load of 3,000 kg and a rear axle load of 6,500 kg, indicating a serious imbalance; cargo size and weight characteristics, generated by the multimodal data processing module; vehicle specifications, with a cargo box size of 10m × 2.5m × 2.5m and a maximum load of 10,000 kg.

[0108] Optimization process: In the initial state, the vehicle's weight distribution was severely biased towards the rear axle, and the center of gravity was too far back. The first optimization step was to place cargo A at the front of the truck bed, increasing the front axle load to 5,500 kg and reducing the rear axle load to 5,000 kg. The second optimization step was to place cargo B in the middle left of the truck bed to optimize the load balance between the front and rear axles and prevent the center of gravity of the cargo from shifting. The third optimization step was to place cargo C in the center of the rear of the truck bed to fill the space gap and improve the utilization rate of the truck bed space.

[0109] Output results: The loading schemes generated by the dynamic loading optimization model include:

[0110] Location of Goods A: (1m, 1m, 0m);

[0111] Location of Goods B: (5m, 0.5m, 0m);

[0112] Location of cargo C: (8m, 1.25m, 0m).

[0113] The optimized load distribution results in a front axle load of 5,500 kg and a rear axle load of 4,500 kg, with the center of gravity located in the middle of the carriage, significantly improving load balance and stability.

[0114] Preferably, the dynamic loading optimization model is simulated dynamically through cellular automata rules. The cell state includes the current cargo position, the vehicle center of gravity position, and the cargo priority. The cell state change rules include the impact of the newly added cargo position on the center of gravity of neighboring cells. The simulation results include loading task data and cargo priority ranking.

[0115] The dynamic loading optimization model uses cellular automata (CA) rules for dynamic evolution simulation to optimize the loading sequence and location of goods. A cellular automaton is a discrete dynamic system composed of a set of rule-driven cells whose state evolves over time. This invention, based on the local interaction characteristics of cellular automata, simulates the impact of adding goods during the loading process on the vehicle's center of gravity, spatial distribution, and priority ranking.

[0116] Cellular state and spatial modeling include: cell state definition, where each cell represents a fixed position in the carriage space. Cellular state includes the following information: current cargo position, indicating whether the cell is occupied and the number of the occupied cargo; vehicle center of gravity position, calculated based on the cargo's position and weight within the carriage; cargo priority, defining the cargo's priority in the loading task to guide the loading sequence. Spatial modeling divides the carriage into several three-dimensional cellular units (…). (Coordinate system), each cell represents a fixed, tiny space within the carriage. The state changes of each cell depend on its own properties and its interactions with neighboring cells.

[0117] The cell state change rules are based on the impact of newly added cargo on the center of gravity, space occupancy, and priority ranking of neighboring cells during the loading process. Specific rules include:

[0118] The center of gravity shift rule states that the position of newly added cargo will affect the overall center of gravity distribution of the vehicle. The system updates the contribution of each cell to the overall center of gravity shift in real time, calculated using the following formula:

[0119]

[0120] in, Indicates the offset of the center of gravity; Indicates the first The weight of the newly added goods; Indicates the first The distance from the newly added cargo to the vehicle's current center of gravity; This indicates the vehicle's current total weight.

[0121] The proximity rule applies: adding new goods occupies space, and its status updates affect the availability of neighboring cells. The system marks available space by detecting the occupancy status of neighboring cells.

[0122] Priority ranking rules: the priority of goods is determined by their volume, weight and the importance of the transportation task, and the priority ranking is dynamically adjusted as the cell state changes.

[0123] During the simulation, the system starts from an initial cellular state and gradually updates the 3D cellular state of the carriage according to cellular rules. The placement of newly added cargo triggers iterations of the cellular rules, ultimately resulting in a global loading scheme. The output includes: loading task data, and the loading position of each cargo item (…). (Coordinates) and order. Cargo priority sorting, the optimized loading order, is used to guide the loading robot to perform tasks.

[0124] The local rules and global evolution characteristics of cellular automata enable real-time responses to changes in the state of newly added cargo, dynamically adjusting loading strategies. Through center-of-gravity offset rules, the system optimizes the vehicle's center-of-gravity distribution, ensuring balanced loading and transportation stability. Proximity influence rules maximize the use of cargo space, avoiding space waste caused by irregularly shaped cargo. Priority sorting rules ensure that important cargo is loaded first, while reducing loading time and improving transportation efficiency.

[0125] In an example, during a logistics task, the system needs to load three items of goods into a vehicle. The characteristics of the goods are as follows:

[0126] Goods A: Weight 1,000 kg, dimensions 1.5m × 1.0m × 0.8m, priority 3;

[0127] Goods B: Weight 800kg, dimensions 1.2m×1.0m×0.6m, priority 1;

[0128] Goods C: Weight 600kg, Dimensions 1.0m×0.8m×0.5m, Priority 2.

[0129] Initial state: The carriage is divided into 10×5×3 cellular units (each unit is 1m³). The current center of gravity is located at the center of the bottom of the carriage. All cell states are empty.

[0130] Simulation process: Step 1, cargo B is loaded first, its volume occupies 1 / 3 of the cell. The center of gravity offset is calculated as follows: (Shift forward). Cell state update, available space in adjacent cells decreases. Second step, cargo C is loaded into the middle of the carriage, its placement position is... The center of gravity shift was adjusted to... (Returning to the central axis). Third step: Cargo A is loaded at the rear of the carriage, its position being... This ensures that the remaining space is maximized. The center of gravity shift is ultimately balanced in the middle of the carriage. ).

[0131] Output: Loading task data, placement coordinates and loading order of goods B, C, and A. Goods priority sorting: B>C>A (optimized based on goods characteristics and importance).

[0132] The loading execution module includes a loading robot, which generates loading task data based on dynamic loading optimization model data and performs cargo loading operations; it monitors changes in the vehicle's center of gravity through inertial sensors and captures load changes during cargo loading through piezoelectric sensors, and adjusts the loading sequence and path in real time in conjunction with a dynamic interactive prediction algorithm to optimize the loading process; the dynamic center of gravity distribution data generated during the loading process and the cargo placement status data together constitute the loading status data.

[0133] The loading robot generates loading task data, including the placement order, location coordinates, and loading path of the goods, based on dynamic loading optimization model data. The robot features a multi-axis robotic arm structure and a flexible gripping device, enabling precise grasping and placement of goods. A path planning algorithm determines the optimal transport trajectory of the goods from their initial position to their final placement location.

[0134] Inertial sensors calculate the vehicle's center of gravity position in real time by measuring changes in the vehicle's acceleration. The monitoring data from the inertial sensors, combined with the vehicle's geometric model, is used to update the vehicle's dynamic center of gravity distribution data to determine the load balance.

[0135] Piezoelectric sensors are mounted at the contact points between the robotic arm and the cargo to capture load changes during cargo loading, specifically detecting whether the cargo weight matches expected data and whether the cargo is correctly positioned. The output of the piezoelectric sensor is directly used as the real-time adjustment input for the dynamic interactive prediction algorithm.

[0136] The dynamic interactive prediction algorithm, based on a recurrent neural network (RNN), predicts the impact of each action during the loading process using multi-step time series data and adjusts the loading strategy in real time. Input data includes dynamic center of gravity distribution data, cargo placement status data, and loading path data. Output results include the current optimal placement position of the cargo, the next loading path, and suggestions for sequence adjustment.

[0137] Throughout the loading process, the system generates dynamic center of gravity distribution data and cargo placement status data using inertial sensors and piezoelectric sensors, respectively. The dynamic center of gravity distribution data reflects the vehicle's current load balance and center of gravity position. The cargo placement status data includes the specific placement position, angle, and fixation status of each piece of cargo. These two types of data together constitute the loading status data, which is updated in real time to support subsequent detection and verification modules.

[0138] Preferred, such as Figure 3As shown, the loading robot in the loading execution module has a six-axis robotic arm structure. Combined with a path planning algorithm, it performs precise grasping and placement operations of goods according to the loading task data. The loading robot monitors the changes in the vehicle's center of gravity in real time through inertial sensors and optimizes the loading path by combining a dynamic interactive prediction algorithm. The dynamic interactive prediction algorithm is based on a recurrent neural network to predict the impact of loading sequence adjustments.

[0139] The loading robot in the loading execution module employs a six-axis robotic arm structure, combining path planning algorithms and dynamic interactive prediction algorithms to achieve precise grasping, placement, and path optimization of goods in the unmanned shipping system. Its core lies in using inertial sensors to monitor changes in the vehicle's center of gravity in real time and utilizing recurrent neural networks (RNNs) to predict the impact of loading path adjustments on the center of gravity distribution and task efficiency, thereby achieving dynamic optimization. Equipped with a six-degree-of-freedom robotic arm, the loading robot can operate flexibly in space, adapting to complex variations in cargo size and loading space.

[0140] Based on the cargo and target locations generated from loading task data, the robotic arm's movement trajectory is calculated through path optimization to ensure efficient cargo handling. Optimization formula:

[0141]

[0142] in, Indicates the total handling time; Indicates the first path The distance between each transport point; This represents the penalty value for changes in angle along the path; This represents the angle weighting factor. During execution, the robotic arm dynamically adjusts its path to avoid obstacles and ensures optimal angle changes to improve loading efficiency.

[0143] The vehicle's acceleration and tilt changes during loading are measured using inertial sensors, and the real-time position of the vehicle's center of gravity is calculated using the vehicle's geometric model. The formula for center of gravity calculation is:

[0144]

[0145] in, Indicates the location of the vehicle's center of gravity; Indicates the first The weight of the goods; Indicates the first The placement location of the goods; This indicates the vehicle's current total weight. The system dynamically assesses load balance using center of gravity change data to ensure that the vehicle's center of gravity remains within the design safety range during loading.

[0146] The dynamic interactive prediction algorithm, based on a recurrent neural network (RNN), analyzes the impact of loading sequence adjustments on loading paths and vehicle center of gravity through time-series prediction. Inputs include real-time monitored dynamic center of gravity data, cargo placement status data, and path planning results. Outputs include the optimized loading path and the adjusted loading sequence. Core prediction formula:

[0147]

[0148] in, This indicates the optimized loading path for the next step. This indicates the current path prediction result; The input data represents the current time step, including data on center of gravity changes from inertial sensors and cargo status data. The dynamic interactive prediction algorithm is characterized by its ability to respond in real-time to changes in external inputs, optimizing overall path planning when cargo loading order is adjusted, ensuring balanced vehicle load and improved loading efficiency.

[0149] Dynamic center of gravity distribution data reflects the vehicle's center of gravity distribution at the current loading progress, used to assess load balance. Cargo placement status data records the placement position, angle, and fixation status of each piece of cargo. These two types of data together constitute loading status data, which is updated to the system in real time, providing accurate input for subsequent processes.

[0150] The combination of a six-axis robotic arm structure and path planning algorithms significantly improves cargo loading efficiency while reducing energy consumption during handling. Inertial sensors monitor changes in the vehicle's center of gravity in real time, and a dynamic interactive prediction algorithm dynamically adjusts the loading sequence to ensure the vehicle maintains stability during loading. Recurrent neural networks predict the impact of loading path adjustments, effectively preventing disordered cargo placement or path conflicts during loading. Loading status data is generated in real time, providing accurate vehicle status information for subsequent detection and verification modules.

[0151] In an example, an unmanned delivery system needs to load the following goods into a vehicle:

[0152] Goods A: Weight 1,000 kg, dimensions 1.8m × 1.2m × 1.0m;

[0153] Goods B: Weight 800kg, dimensions 1.5m × 1.0m × 0.8m;

[0154] Goods C: Weight 1,200kg, dimensions 2.0m×1.5m×1.2m.

[0155] The task data generated by the dynamic loading optimization model includes:

[0156] Cargo A is placed at the front of the carriage. );

[0157] Cargo B is placed in the middle of the carriage. );

[0158] Cargo C is placed at the rear of the carriage. ).

[0159] The loading robot transports goods one by one to designated locations according to a path planning algorithm-optimized trajectory: the transport path for goods A is planned as a straight line to reduce turning time; goods B avoids the trajectory of already loaded goods A midway, optimizing the transport angle to save path length; goods C avoids obstacles at the rear of the vehicle through dynamic path adjustment. During the transport process, inertial sensors detect that the vehicle's center of gravity gradually shifts from the front axle to the center, and the system dynamically adjusts the placement positions of goods B and C to ensure a stable center of gravity.

[0160] Dynamic center of gravity distribution data: the front axle load accounts for 40% of the total weight, the rear axle load accounts for 60%, and the center of gravity is located on the vehicle's centerline; cargo placement data includes the coordinates, placement angle, and fixed status of each piece of cargo.

[0161] Preferably, the dynamic interactive prediction algorithm combines cargo size feature data and vehicle loading space model data to recalculate loading sequence and path optimization parameters after each cargo placement operation, generating real-time adjusted loading task data.

[0162] The dynamic interactive prediction algorithm is an optimization method based on cargo size characteristic data and vehicle loading space model data in an intelligent weighing unmanned shipping system. It aims to recalculate loading sequence and path optimization parameters after each cargo placement operation and generate real-time adjusted loading task data. Through this dynamic adjustment mechanism, the system can adapt to changes in the actual loading process (such as space utilization and center of gravity shift) and optimize the overall loading plan.

[0163] Cargo size feature data, including the length, width, height, and spatial shape of the cargo, originates from the multimodal data processing module. Vehicle loading space model data, obtained through 3D modeling technology, divides the vehicle's available loading space into regular grid cells, labeling the availability status of each cell (available, occupied, unavailable). A dynamic interactive prediction algorithm matches the cargo size feature data with available cells in the current loading space model to select the optimal placement area. Matching calculation formula:

[0164]

[0165] in, Indicates the matching score; Indicates the first The volume of each available space unit; Indicates the current volume of the goods; This indicates the total number of available space units.

[0166] The dynamic interactive prediction algorithm is based on a recurrent neural network (RNN). It combines the size characteristics of the cargo with real-time loading status data generated after the placement operation to recalculate the loading sequence and path optimization parameters.

[0167] After each cargo placement, the algorithm updates the loading space model in real time, marking the current cargo occupancy status and the availability of adjacent areas. Time series prediction is used to analyze the impact of the current operation on the loading paths and order of subsequent cargo. Optimization formula:

[0168]

[0169] in, This indicates the optimized loading task data for the next step. Indicates the current task status; This indicates the current input data (including cargo size characteristics and loading space model data).

[0170] Loading sequence optimization dynamically adjusts the loading order based on cargo priority, current placement, and remaining available space, prioritizing larger cargo to maximize space utilization. Route optimization parameters recalculate the optimal parameters for the loading path, avoiding unnecessary repetitive movements and complex routes.

[0171] After each cargo placement operation, the system dynamically adjusts the subsequent loading sequence and route planning to ensure the overall loading process is optimized. By updating the loading space model and center of gravity position in real time, it ensures the even distribution of vehicle load and reduces transportation risks. Precise matching of cargo dimensions with the loading space significantly improves the utilization efficiency of the cargo compartment, avoiding a decrease in loading capacity due to wasted space. Dynamic adjustment of route optimization parameters reduces ineffective actions during the loading process, improving overall operational efficiency.

[0172] In an example, during a logistics loading task, the system needs to load the following goods into the vehicle compartment:

[0173] Goods A: Weight 1,200kg, dimensions 1.8m × 1.2m × 1.0m;

[0174] Goods B: Weight 800kg, dimensions 1.5m × 1.0m × 0.8m;

[0175] Goods C: Weight 1,000 kg, dimensions 2.0m × 1.5m × 1.2m.

[0176] Initial loading task data generation, and initial calculation results of the dynamic loading optimization model:

[0177] Goods A should be placed at the front of the carriage, at the following location: );

[0178] Goods B is placed in the middle left of the carriage, at a position of ( );

[0179] Cargo C is placed at the rear of the carriage, at the location of ( ).

[0180] Dynamic adjustment after cargo A is placed: After cargo A is placed, the system updates the loading space model, marking the area occupied by cargo A and the availability status of adjacent units; the dynamic interactive prediction algorithm analyzes the current center of gravity shift, detecting that the front axle load has increased to 40%, the rear axle load has decreased to 60%, and the center of gravity is slightly forward. The system adjusts the loading sequence, placing cargo B in the middle and slightly rear of the carriage to balance the center of gravity distribution.

[0181] Path optimization after placing cargo B: After placing cargo B, the path optimization parameters are recalculated, and the redundant movement parts contained in the original path are removed; the total path length is reduced by 20%.

[0182] The loading task data was adjusted in real time: the final loading order was adjusted to: Cargo A > Cargo C > Cargo B. The loading path optimization results show that the total path length was reduced and the handling time was shortened by 15%.

[0183] like Figure 4 As shown, the detection and verification module is used to collect vehicle detection data after loading is completed, generate verification results based on the comparison between the detection data and the loading status data, and generate an electronic shipping document when the verification is passed.

[0184] Preferably, the dynamic detection device in the detection and verification module collects the actual load distribution data of the vehicle through a pressure sensor, and generates the actual center of gravity data and vibration mode data of the vehicle by combining it with an inertial sensor and a vibration sensor. The comparison algorithm calculates the deviation range based on the vehicle design load model and loading state data, and generates verification results based on the deviation range.

[0185] The dynamic detection device consists of pressure sensors, inertial sensors, and vibration sensors. These sensors work together to collect real-time data from the vehicle. The pressure sensors collect pressure data at the tire contact points to calculate the vehicle's actual load distribution. The load distribution calculation formula is as follows:

[0186]

[0187] in, Indicates the vehicle's current total weight; Indicates the first Pressure values ​​collected by a pressure sensor; Indicates the first The effective contact area of ​​each pressure node; This indicates the total number of pressure sensor nodes.

[0188] Inertial sensors monitor the vehicle's acceleration changes in real time and calculate the actual center of gravity data based on the vehicle's geometric model. Vibration sensors collect the vibration frequency and amplitude of the vehicle during operation, generating vibration pattern data to determine the fixation and stability of the cargo.

[0189] The deviation comparison algorithm, based on the vehicle design load model and loading status data, calculates the deviation range between the vehicle's actual and ideal states. Input data includes: actual load distribution data, actual center of gravity data, and vibration mode data; the vehicle design load model, including the vehicle's maximum load limit, front and rear axle load balancing parameters, and cargo securing requirements; and loading status data, including cargo placement, load distribution, and the vehicle's current center of gravity position. The deviation calculation formula is as follows:

[0190]

[0191] in, Indicates deviation of the center of gravity; Indicates the current position of the vehicle's center of gravity; This represents the ideal center of gravity position in the design load model. Deviation range generation involves setting a deviation threshold based on vehicle design requirements (e.g., center of gravity offset not exceeding 10%). If the calculation result exceeds the threshold, a verification failure result is generated.

[0192] Based on the calculation results of the deviation comparison algorithm, a verification report is generated, including: load distribution deviation results (whether they meet the balance requirements of the vehicle design load model); center of gravity deviation results (whether the current center of gravity meets the design center of gravity offset limit); and vibration mode deviation results (whether the vibration is within the safe range).

[0193] The actual load, center of gravity, and vibration status of vehicles can be comprehensively monitored before transportation, ensuring that the load status of each vehicle meets safety requirements. Deviation comparisons are performed using loading status data, and adjustment suggestions are generated promptly when problems are detected to optimize loading results. Vibration pattern data is used to determine the stability of cargo, preventing safety accidents caused by loose cargo or shifted center of gravity. The verification results generated using deviation ranges provide high-precision basic data support for the subsequent generation of electronic shipping documents.

[0194] In an example, a logistics company uses an intelligent weighing and unmanned shipping system to inspect a loaded truck. The goods include:

[0195] Goods A: Weight 1,200kg, Location (1m, 1m, 0m);

[0196] Goods B: Weight 800kg, Location (5m, 1m, 0m);

[0197] Cargo C: Weight 1,000 kg, Location (9m, 1.5m, 0m).

[0198] Dynamic detection data acquisition: The pressure sensor collected a front axle load of 5,000 kg, a rear axle load of 6,000 kg, and a total weight of 11,000 kg; the inertial sensor calculated the current center of gravity position to be slightly rear of the middle of the vehicle body. The vibration sensor detected a vibration frequency of 8 Hz and an amplitude of 0.02 m.

[0199] Deviation comparison calculation: Load distribution deviation: According to the vehicle design load model, the front and rear axle load ratio should be 45:55, and the actual deviation is 5%, which passes the verification. Center of gravity deviation: The ideal center of gravity position is the middle of the carriage (…). The actual deviation is The verification passed. Vibration mode deviation: the detected vibration frequency and amplitude are within the vehicle's permissible range (frequency <10Hz, amplitude <0.03m), verification passed.

[0200] The verification results are generated, including: Total weight: 11,000 kg, conforming to the vehicle design load model; Front and rear axle load ratio: 45:55, deviation 5%, verification passed; Center of gravity position: The deviation was 8.3%, and the verification passed; the vibration mode was safe, and the verification passed. The results showed that the vehicle load was balanced, the center of gravity was stable, and the vibration was normal. An electronic shipping order was generated, and the vehicle was allowed to be shipped.

[0201] Preferably, the electronic waybill includes vehicle information, cargo information, shipping time, verification results, and deviation range evaluation indicators. The electronic waybill is synchronized to the transportation management system through a digital twin entity for status tracking of subsequent logistics operations.

[0202] The data in the electronic shipping order comes from the outputs of the detection and verification module and the loading execution module, specifically including:

[0203] Vehicle information: Basic data such as vehicle registration number, vehicle type, and maximum load;

[0204] Cargo information: Cargo number, specifications (length, width, height), weight, and material properties;

[0205] Shipment time: The shipment confirmation time after the vehicle passes the inspection module verification;

[0206] Verification results: Whether the vehicle's loading status meets the design requirements, including indicators such as load distribution balance, center of gravity offset range, and vibration mode;

[0207] Deviation range evaluation index: The results of various deviations calculated by comparison algorithm, such as front and rear axle load ratio deviation, center of gravity position deviation, and vibration amplitude, are within the safe range.

[0208] Data integration and document generation: The system uses the data integration module to compile the above data into a standardized electronic document format (such as XML or JSON) to ensure the integrity and parsability of the document content.

[0209] A digital twin is a dynamic model of a vehicle and cargo in a virtual environment, reflecting their physical state and operational processes in real time. The content of the electronic waybill is synchronized to the digital twin via edge computing nodes. The digital twin updates its relevant status, including: the vehicle's current status (location, load distribution); cargo loading status (list of loaded cargo and their specific placement); verification results and deviation indicators.

[0210] The synchronized digital twin will connect to the transportation management system via a network interface to enable real-time tracking and sharing of logistics status.

[0211] Through electronic waybills and digital twin entities, logistics managers can monitor the transportation status of vehicles and goods in real time, tracking information including: the current location and progress of the vehicles; whether there are any abnormalities in the loading status (such as loose cargo or shift in the center of gravity); and whether dynamic adjustments to the transportation plan are needed (such as reallocating goods or changing the transportation route).

[0212] When the digital twin detects a deviation that exceeds the safe range (such as excessive center of gravity shift), the system will automatically send an alarm message to the transportation management system and suggest adjustments to the plan.

[0213] Electronic waybills integrate comprehensive information on vehicles, goods, verification results, and deviation indicators, ensuring transparency of logistics task information. The synchronization function of digital twin entities ensures real-time updates of dynamic data during transportation, providing end-to-end traceability support for logistics operations. Deviation range evaluation indicators provide a safety guarantee for logistics tasks and can provide early warnings of potential risks. Through integration with the transportation management system, managers can access the status of vehicles and goods at any time, improving logistics management efficiency.

[0214] In an example, a logistics company uses an intelligent weighing unmanned shipping system to transport a batch of goods for a customer. The electronic shipping order generation and synchronization process for the shipping task is as follows:

[0215] Electronic shipping document data integration:

[0216] Vehicle information: Vehicle number is "L001", vehicle type is "12T flatbed truck", maximum load is 12,000 kg;

[0217] Cargo information: Cargo A (weight 1,000kg, dimensions 1.5m×1.2m×1.0m, material is "polyethylene"), Cargo B (weight 800kg, dimensions 1.2m×1.0m×0.8m, material is "metal");

[0218] Shipping time: 10:30 AM, January 15, 2024;

[0219] Verification result:

[0220] Load distribution balance: The front and rear axle load ratio is 45:55, with a deviation of 5%, which meets the design requirements;

[0221] Center of gravity offset range: deviation 8%, within the safe range;

[0222] Vibration mode: frequency 7Hz, amplitude 0.015m, within the safe range;

[0223] Deviation range evaluation index: All deviation indexes are within the set threshold range, and the verification is passed.

[0224] Electronic shipping order generation and synchronization: The system integrates the above data to generate an electronic shipping order and synchronizes it to the digital twin entity. The digital twin entity is updated as follows:

[0225] Current load: Total weight 1,800 kg, front axle load 800 kg, rear axle load 1,000 kg;

[0226] Cargo status: Cargo A and cargo B have been loaded and placed in the front and middle of the carriage, respectively;

[0227] Verification result: All indicators passed.

[0228] During transportation, logistics managers monitor vehicle status in real time through the transportation management system. The digital twin entity shows that the vehicle has reached the highway entrance and is currently in normal condition; if the vibration amplitude of the cargo compartment increases to 0.03m (exceeding the safe range) during transportation, the system automatically alarms and suggests slowing down to prevent the cargo from becoming loose.

[0229] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0230] The above are merely embodiments of this application and are not intended to limit the scope of 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 scope of the claims of this application.

Claims

1. An unmanned shipping system based on intelligent weighing, characterized in that, Comprise: A multi-modal data processing module for collecting multi-modal data including vehicle feature data, driver feature data and cargo feature data; Quantum encoding of the multi-modal data, mapping multi-modal data to high-dimensional space using kernel function method, and extracting core feature correlation between data through high-dimensional correlation analysis to generate correlation feature data, and combining task information to generate digital twin entity data; Distributed weighing and optimization module for calculating weight distribution model through vehicle contact pressure data, and combining vehicle specification information and cargo feature data to generate dynamic loading optimization model using reinforcement learning optimization algorithm; the dynamic loading optimization model is simulated by dynamic evolution through cellular automata rules, and the cellular state includes current cargo position, vehicle center of gravity position and cargo priority, and the cellular state change rule includes the influence of new cargo position on adjacent cellular center of gravity, and the simulation result includes loading task data and cargo priority sorting; Cellular state and space modeling includes: cellular state definition, each cell represents a fixed position in the vehicle compartment space, and the cellular state includes the following information: current cargo position, used to identify whether the cell is occupied and the number of occupied cargo; vehicle center of gravity position, used to calculate the vehicle center of gravity coordinates according to the position and weight of the cargo in the vehicle compartment; cargo priority, used to define the priority of the cargo in the loading task and guide the loading sequence; space modeling, used to divide the vehicle compartment into a plurality of three-dimensional cellular units, each cellular unit represents a fixed micro space in the vehicle compartment; the state change of each cell depends on its own characteristics and interaction with adjacent cells; The cellular state change rule is based on the influence of new cargo on the center of gravity of adjacent cells, space occupation and priority sorting in the loading process; the change rule includes: Center of gravity offset rule, the position of the new cargo will affect the overall center of gravity distribution of the vehicle; the system updates the contribution of each cell to the overall center of gravity offset in real time, and the calculation formula is: in, Indicates the offset of the center of gravity; Indicates the first The weight of the newly added goods; This indicates the total quantity of newly added goods; Indicates the first The distance from the newly added cargo to the vehicle's current center of gravity; Indicates the vehicle's current total weight; Adjacent influence rule, the new cargo will occupy space, and its state update will affect the availability of adjacent cells; the system marks the available space by detecting the occupation of adjacent cells; Priority sorting rule, the priority of the cargo is determined by its volume, weight and importance of the transportation task, and the priority sorting is dynamically adjusted with the change of cellular state; In the simulation process, the system starts from the initial cellular state, and gradually updates the three-dimensional cellular state of the vehicle compartment according to the cellular rules; the placement position of the new cargo will trigger the iteration of the cellular rules, and the global loading scheme is finally obtained; the output result includes: loading task data, the loading position and sequence of each cargo; cargo priority sorting, the optimized loading sequence, used to guide the loading robot to execute the task; The loading execution module includes a loading robot for generating loading task data according to dynamic loading optimization model data and performing cargo loading operations; the vehicle's center of gravity changes are monitored through an inertial sensor, and the load changes during the cargo loading process are captured through a piezoelectric sensor, and the loading sequence and path are adjusted in real time to optimize the loading process in combination with a dynamic interactive prediction algorithm; the dynamic center of gravity distribution data generated during the loading process and the cargo placement state data jointly constitute the loading state data; The detection and verification module is used for collecting detection data of the vehicle after the loading is completed, generating a verification result based on the comparison of the detection data and the loading state data, and generating an electronic delivery single when the verification is passed.

2. The smart-weighing-based unattended delivery system according to claim 1, characterized in that, The spectral scanning device in the multi-modal data processing module emits and receives spectral data reflected by the vehicle surface through a multi-frequency band light source, and generates vehicle spectral feature data in combination with a preset spectral feature matching algorithm; The face recognition camera classifies and extracts the facial features of the driver through a convolutional neural network algorithm to generate driver facial feature data; the three-dimensional laser scanner obtains three-dimensional size data of the cargo surface through laser beam ranging technology to generate cargo size feature data.

3. The smart-weighing-based unattended delivery system according to claim 2, characterized in that, The chemical sensor in the multi-modal data processing module detects the molecular composition of the cargo material through gas-phase molecular analysis, generates material characteristic description data of the cargo in combination with a material feature model matching algorithm stored in the database, and inputs the data as part of the cargo feature data into the multi-modal data processing module.

4. The smart-weighing-based unattended delivery system according to claim 1, characterized in that, The pressure sensor array in the distributed weighing and optimization module is installed in a grid-like manner in the vehicle driving path, generates vehicle weight distribution data by collecting pressure data of the vehicle tire contact points in real time, and combines a vehicle axle load distribution model.

5. The smart-weighing-based unattended shipping system according to claim 1, characterized in that, The distributed weighing and optimization module uses a deep Q network structure in a reinforcement learning optimization algorithm to input vehicle weight distribution data, three-dimensional size feature data of the cargo, and vehicle specification information, and outputs a dynamic loading optimization model of the vehicle, wherein the training target of the reinforcement learning optimization algorithm includes minimization of loading balance, center of gravity stability, and loading time efficiency.

6. The smart-weighing-based unattended shipping system according to claim 1, characterized in that, The loading robot in the loading execution module has a six-axis mechanical arm structure, combines a path planning algorithm, and performs precise grabbing and placing operations of the cargo according to the loading task data; the loading robot monitors the center of gravity changes of the vehicle in real time through an inertial sensor, and optimizes the loading path in combination with a dynamic interactive prediction algorithm, wherein the dynamic interactive prediction algorithm predicts the influence of loading sequence adjustment based on a recurrent neural network.

7. The smart-weighing-based unattended delivery system according to claim 6, characterized in that, The dynamic interactive prediction algorithm combines cargo size feature data and vehicle loading space model data to recalculate loading sequence and path optimization parameters after each cargo placement operation, and generates real-time adjusted loading task data.

8. The smart-weighing-based unattended shipping system according to claim 1, characterized in that, The dynamic detection device in the detection and verification module collects actual load distribution data of the vehicle through a pressure sensor, and generates actual center of gravity data and vibration mode data of the vehicle in combination with an inertial sensor and a vibration sensor; a comparison algorithm calculates a deviation range based on a vehicle design load model and loading state data, and generates a verification result based on the deviation range.

9. The smart-weighing-based unattended shipping system according to claim 1, characterized in that, The electronic shipping order includes vehicle information, cargo information, shipping time, verification result and deviation range evaluation index, and the electronic shipping order is synchronized to the transportation management system through a digital twin entity, for state tracking of subsequent logistics operations.

Images