An edge-computing-based loading platform autonomous decision-making and real-time communication system and method

Abstract

This invention discloses an edge computing-based autonomous decision-making and real-time communication system and method for loading depots, relating to the field of railway freight automation. The system includes: deploying edge computing nodes to construct an end-to-end distributed processing architecture, collecting real-time data and performing localized processing; utilizing an autonomous decision-making module to construct a closed-loop mechanism of "state awareness - strategy reasoning - instruction output" based on uploaded multi-source data, autonomously generating material unloading control instructions and multi-device collaborative strategies; and employing a real-time communication protocol integrating Time-Sensitive Networking (TSN) and 5G URLLC technologies to ensure real-time communication between edge computing nodes, between edge computing nodes and the autonomous decision-making module, and when equipment executes instructions, achieving synchronization between locomotive arrival and material unloading initiation, and seamless connection between leveling machine operation and material unloading completion. This invention can improve the unmanned, highly collaborative, and low-latency capabilities of loading operations, thereby better adapting to the needs of the "heavy-load" transportation mode in railway freight transport.

Description

Technical Field

[0001] This invention relates to the field of railway freight automation technology, and more specifically to a loading terminal autonomous decision-making and real-time communication system and method based on edge computing. Background Technology

[0002] Railway freight collection and distribution, as a crucial link connecting railway trunk lines and terminal freight yards, undertakes the task of organizing the entire process of "gathering-trunk line transportation-dispersion" of goods. The "return-return" model refers to loading goods onto freight cars and departing from the source, unloading them at the destination, and then reloading them with full loads on the return trip, forming an efficient closed loop. This significantly reduces logistics costs and is the core path for cost reduction and efficiency improvement in railway bulk freight transportation.

[0003] The efficient operation of the "repeated loading and unloading" model has placed demands on loading operations to be "unmanned, highly collaborative, and low-latency." This upgraded demand has prompted a transformation in loading technology from "traditional manual-dominated" to "intelligent and collaborative." As the core equipment for loading operations, loading towers improve efficiency and accuracy through functions such as high-level storage and continuous conveying, precise metering and loading, and semi-automated operation. However, despite significant advancements in loading towers compared to traditional methods, they still face significant limitations under the new demands for "unmanned operation, high collaboration, and low latency," specifically manifested in the following ways:

[0004] 1) Existing loading terminal systems still rely on manual intervention for unmanned operation and cannot achieve full-process automated control;

[0005] 2) Insufficient data collaboration capabilities among multiple systems lead to severe information silos, affecting the overall collaboration of operations;

[0006] 3) The delay in loading operations remains high, which cannot meet the needs of rapid turnover.

[0007] Therefore, how to improve the unmanned, highly collaborative, and low-latency capabilities of loading operations to better adapt to the needs of the "heavy-loaded" mode of railway collection and distribution is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides an autonomous decision-making and real-time communication system and method for loading depots based on edge computing, which solves the problems existing in the background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for autonomous decision-making and real-time communication in a loading terminal based on edge computing includes the following steps:

[0011] S1. Deploy edge computing nodes at the loading tower, unmanned locomotives, and intelligent leveling machines to build an "end-to-end" distributed processing architecture; collect real-time data through edge computing nodes and perform localized processing, then upload the processed multi-source data to the autonomous decision-making module;

[0012] S2. Utilizing the autonomous decision-making module, a closed-loop mechanism of "state perception - strategy reasoning - instruction output" is constructed based on the uploaded multi-source data to autonomously generate material dropping control instructions and multi-device collaborative strategies.

[0013] S3. It adopts a real-time communication protocol that integrates Time-Sensitive Networking (TSN) and 5G URLLC technology to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when equipment executes instructions, so as to realize the synchronization of locomotive arrival and material dropping start, and the connection between the leveling machine action and the end of material dropping.

[0014] Optionally, in S1, an "end-to-end" distributed processing architecture is constructed, specifically as follows:

[0015] The loading tower is equipped with a first edge computing node with multi-interface data acquisition function. The first edge computing node is connected to the loading tower's material unloading control unit, material level sensor, and weighing device through an industrial bus and is deployed in the loading tower's control cabinet.

[0016] The unmanned locomotive is equipped with a second edge computing node that has GNSS positioning and vehicle bus adaptation functions. The second edge computing node is connected to the locomotive's automatic driving system and carriage positioning sensors via vehicle Ethernet and is deployed in the electrical cabinet of the unmanned locomotive.

[0017] A third edge computing node with motion control signal processing function is configured for the intelligent leveling machine. The third edge computing node is connected to the leveling machine's actuator and attitude sensor through a wireless communication module and is deployed in the protective cabin of the intelligent leveling machine.

[0018] Physical connections between edge computing nodes are established through a fiber optic ring network, and a distributed consensus algorithm is used to achieve clock synchronization and data sharding storage between nodes, forming an end-to-end redundant communication link.

[0019] Optionally, in S1, real-time data is collected and processed locally, specifically as follows:

[0020] Edge computing nodes are used to collect data on material flow rate, material level in silos, and gate opening of the loading tower; real-time location, speed, carriage number and size data of unmanned locomotives; and working position, actuator posture and drive motor current data of intelligent leveling machines.

[0021] The collected real-time data is filtered, denoised, timestamped, and normalized. Key features are extracted, and the abnormal state of the equipment is judged based on a preset threshold. Structured data containing the equipment operating status identifier is generated and uploaded to the autonomous decision-making module.

[0022] Optionally, the specific steps of S2 are as follows:

[0023] Spatiotemporal fusion of multi-source data is performed, and parameters related to equipment operating status and working environment are aggregated through feature weighting algorithms to construct a multi-dimensional state vector;

[0024] A deep reinforcement learning algorithm is adopted, with the dual objective functions of minimizing energy consumption and maximizing loading efficiency. Iterative training is carried out in a variety of preset scenarios to generate a dynamic material loading allocation strategy and equipment action timing plan.

[0025] The dynamic material feeding allocation strategy is converted into standardized control commands, which are then sent to the corresponding devices via the S3 real-time communication protocol. The system also receives device execution feedback data and updates the state vector to complete the closed-loop iteration.

[0026] Optionally, the specific methods for generating dynamic material feeding allocation strategies and equipment action timing planning are as follows:

[0027] Based on the current material characteristics, the location of the carriage, and the loading progress, the opening and closing timing and flow distribution coefficient of each material discharge port are output through the strategy network to form a material discharge scheme containing multiple control parameters.

[0028] The loading operation is decomposed into discrete control cycles of 100ms by using a time window sliding algorithm, generating a sequence of coordinated control instructions that includes locomotive travel speed curve, leveler insertion depth and timing, and material dropping system start and stop delay.

[0029] Optionally, the specific steps of S3 are as follows:

[0030] Edge computing nodes of fixed equipment build wired links through converged time-sensitive network switches, while edge computing nodes and autonomous decision-making modules of mobile devices access the network through 5G URLLC.

[0031] Nanosecond-level clock alignment is performed on devices within the converged time-sensitive network to synchronize the 5G URLLC terminal time to the core network reference clock;

[0032] Define a standardized instruction format with timestamps. Edge computing nodes and autonomous decision-making modules interact with each other through a real-time communication protocol to provide device status feedback. Dynamically calibrate the material dropping start delay and the leveling machine action trigger point to ensure that the response time from the locomotive's arrival to the material dropping start is less than 50 milliseconds and the connection error from the end of material dropping to the leveling machine action is less than 20 milliseconds.

[0033] A system for implementing the edge computing-based autonomous decision-making and real-time communication method for loading depots as described above, comprising:

[0034] Edge computing node clusters are deployed in loading towers, unmanned locomotives, and intelligent leveling machines to form an "end-to-end" distributed processing architecture. They are used to collect real-time data and perform local processing, and upload key decision results to the autonomous decision-making module.

[0035] The autonomous decision-making module is used to construct a closed-loop mechanism of "state perception-strategy reasoning-instruction output" based on multi-source data, and autonomously generate material feeding control instructions and multi-device collaborative strategies.

[0036] The real-time communication module adopts a real-time communication protocol that integrates time-sensitive networking and 5G URLLC technology to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when devices execute commands.

[0037] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a loading terminal autonomous decision-making and real-time communication system and method based on edge computing, which has the following beneficial effects:

[0038] 1) Improved operational efficiency: The autonomous decision-making module achieves millisecond-level response based on reinforcement learning algorithms; the dynamic material distribution strategy automatically adjusts the material dropping speed according to material characteristics; the real-time communication protocol ensures improved synchronization accuracy between locomotive arrival and material dropping start, and reduces the connection error between leveling machine action and material dropping completion; more than 80% of the data is processed through edge computing nodes, reducing reliance on the cloud.

[0039] 2) Cost optimization: This invention realizes unmanned loading operations, reduces on-site operators, and lowers maintenance costs; the dual-objective optimization algorithm balances energy consumption and efficiency, and the self-identification of materials and self-adaptation of vehicle models reduces loading errors and lowers material loss rate;

[0040] 3) Enhanced reliability and security: The redundant transmission mechanism integrating TSN and 5G URLLC ensures the reliability of critical control command transmission and avoids timing conflicts in multi-device collaboration; edge computing nodes monitor device status in real time, improving the accuracy of fault prediction; and the localized decision-making mechanism can maintain autonomous operation capability even when the network is interrupted, ensuring operation continuity.

[0041] Therefore, the overall technical solution can improve loading efficiency, reduce labor costs, and improve the accuracy of fault prediction, solving problems such as communication delay, collaboration conflict, and insufficient intelligent decision-making in existing technologies. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0043] Figure 1 A flowchart of the edge computing-based autonomous decision-making and real-time communication method for loading towers provided by the present invention;

[0044] Figure 2 This is a schematic diagram of the "end-to-end" distributed processing architecture provided by the present invention;

[0045] Figure 3 A schematic diagram of the structure of the intelligent decision-making model provided by this invention;

[0046] Figure 4 The diagram shows the structure of the edge computing-based autonomous decision-making and real-time communication system for loading depots provided by this invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Under the "return-on-return" model of railway freight transport, with the continuous increase in the volume of bulk commodity transportation, loading operations have an urgent need for intelligent upgrades, requiring "unmanned operation, high collaboration, and low latency." Addressing the numerous problems exposed by existing technologies at the communication, collaboration, and decision-making levels, this invention discloses a method for autonomous decision-making and real-time communication in loading terminals based on edge computing. Figure 1 As shown, it includes the following steps:

[0049] S1. Deploy edge computing nodes at the loading tower, unmanned locomotive, and intelligent leveling machine to construct an "end-to-end" distributed processing architecture; collect real-time data through edge computing nodes and perform local processing, then upload the processed multi-source data to the autonomous decision-making module; in this embodiment, each edge computing node has data acquisition, data processing, and local decision-making functions, and can independently process the real-time data of the device it is located in, achieve local decision-making as a backup, ensure the continuity of operation, reduce the amount of data transmission, and maximize overall efficiency;

[0050] S2. Utilizing the autonomous decision-making module, a closed-loop mechanism of "state perception - strategy reasoning - instruction output" is constructed based on the uploaded multi-source data. The module autonomously generates material dropping control instructions and multi-device collaborative strategies. For example, when it is detected that the carriage is small and the material has good flowability, it generates instructions for a lower material dropping speed and a more concentrated material dropping position, and coordinates the leveling machine to perform appropriate leveling operations during the material dropping process.

[0051] S3 employs a real-time communication protocol that integrates Time-Sensitive Networking (TSN) and 5G URLLC technologies to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when equipment executes commands. This enables synchronization between locomotive arrival and material unloading start, and seamless connection between leveling machine operation and material unloading completion. TSN provides deterministic transmission latency and jitter, guaranteeing data transmission with high real-time requirements; while 5G URLLC technology offers ultra-reliable and ultra-low latency communication capabilities, further enhancing communication reliability and timeliness.

[0052] Furthermore, in S1, refer to Figure 2 The architecture for constructing an "end-to-end" distributed processing system is as follows:

[0053] The loading tower is equipped with a first edge computing node with multi-interface data acquisition function. The first edge computing node is connected to the loading tower's material unloading control unit, material level sensor, and weighing device through an industrial bus and is deployed in the loading tower's control cabinet.

[0054] The unmanned locomotive is equipped with a second edge computing node that has GNSS positioning and vehicle bus adaptation functions. The second edge computing node is connected to the locomotive's automatic driving system and carriage positioning sensors via vehicle Ethernet and is deployed in the electrical cabinet of the unmanned locomotive.

[0055] A third edge computing node with motion control signal processing function is configured for the intelligent leveling machine. The third edge computing node is connected to the leveling machine's actuator and attitude sensor through a wireless communication module and is deployed in the protective cabin of the intelligent leveling machine.

[0056] Physical connections between edge computing nodes are established through a fiber optic ring network, and a distributed consensus algorithm is used to achieve clock synchronization and data sharding storage between nodes, forming an end-to-end redundant communication link.

[0057] Furthermore, in S1, real-time data is collected and processed locally, specifically as follows:

[0058] Edge computing nodes are used to collect data on material flow rate, material level in silos, and gate opening of the loading tower; real-time location, speed, carriage number and size data of unmanned locomotives; and working position, actuator posture and drive motor current data of intelligent leveling machines.

[0059] The collected real-time data is filtered, denoised, timestamped, and normalized to extract key features such as material flow fluctuation characteristics, locomotive positioning deviation, and leveler load change rate. Based on preset thresholds, abnormal equipment status is determined, and structured data containing equipment operating status identifiers is generated and uploaded to the autonomous decision-making module.

[0060] The processing of real-time data is a core component of edge computing nodes' localized decision-making, and its specific content and steps are as follows:

[0061] (1) Data preprocessing

[0062] a. Filtering and denoising: For the material level sensor data of the loading tower, the Kalman filter algorithm is used to eliminate high-frequency fluctuations in material level caused by mechanical vibration and retain the true trend of material level change; for the GNSS positioning data of the locomotive, the sliding window mean filter is used to remove the positioning jump caused by multipath effect and improve the smoothness of the position data; for the current signal of the leveling machine, the wavelet threshold denoising algorithm is applied to filter the electromagnetic interference in the motor operation and extract the effective load current characteristics.

[0063] b. Timestamp Alignment: Based on the IEEE 1588v2 precision time protocol, the data of each device is uniformly calibrated to the master clock of the edge computing node; for asynchronously acquired data, the time breakpoint is supplemented by linear interpolation to form isochronous sequence data with 10ms interval; a time association table is established to realize the spatiotemporal matching of "locomotive position - material discharge flow - material leveling action".

[0064] c. Format normalization: Convert data of different units into dimensionless values ​​to complete the standardization of physical quantities; convert raw sensor data into JSON format containing "device ID-timestamp-feature value-confidence level" to complete the unification of data structure; use the 3σ criterion to identify and replace data that exceeds the reasonable range to complete outlier handling.

[0065] (2) Key Feature Extraction

[0066] a. Material flow fluctuation characteristics: Calculate the standard deviation and coefficient of variation of the flow rate within the sliding window to identify sudden changes in flow rate caused by material agglomeration; extract the main frequency component of the flow rate signal through Fourier transform to determine the stability of the conveyor belt operation; establish a flow accumulation curve to predict the remaining time to reach the target loading capacity.

[0067] b. Locomotive positioning deviation value: Calculate the lateral and longitudinal deviations between the real-time position and the theoretical loading trajectory, extract the deviation change rate, and predict whether the locomotive will deviate from the loading area; combine the car length parameter to calculate the relative positional relationship between the material discharge port and the car.

[0068] c. Load change rate of leveling machine: Calculate the load change per unit time based on the current signal to determine the material bulk density; extract the location characteristics of the load peak to identify the high-piling area of ​​the material in the carriage; calculate the load mean and standard deviation to evaluate the uniformity of the leveling operation.

[0069] Based on the above processing flow, this embodiment compresses the amount of original data, improves the signal-to-noise ratio of feature data, and provides high-quality data input for subsequent autonomous decision-making.

[0070] In this embodiment, the edge computing node performs localized processing on more than 80% of the collected real-time data. For example, it analyzes the locomotive location data to determine whether the locomotive has reached the designated loading position, and only uploads the key decision result "the locomotive has reached the designated position" to the cloud.

[0071] Furthermore, in the technical solution of this embodiment, the specific steps of S2 are as follows:

[0072] Spatiotemporal fusion of multi-source data is performed, and parameters related to equipment operating status and working environment are aggregated through feature weighting algorithms to construct a multi-dimensional state vector; specifically, the relevant parameters include locomotive position accuracy, car occupancy rate, material humidity, etc.

[0073] A deep reinforcement learning algorithm is adopted, with the dual objective functions of minimizing energy consumption and maximizing loading efficiency. It is iteratively trained in various preset scenarios such as material types, vehicle models, and working conditions to generate a dynamic material distribution strategy and equipment action timing plan. In this embodiment, the reinforcement learning algorithm can achieve material self-identification and vehicle model self-adaptation by discretely simulating various scenarios such as different material dropping speeds, material dropping positions, and leveling machine working modes, thereby achieving a balance between energy consumption and loading efficiency.

[0074] The dynamic material feeding allocation strategy is converted into standardized control commands, which are then sent to the corresponding devices via the S3 real-time communication protocol. The system also receives device execution feedback data and updates the state vector to complete the closed-loop iteration.

[0075] In this embodiment, regarding the deep reinforcement learning algorithm, the Proximal Policy Optimization (PPO) algorithm can be used to construct the intelligent decision-making model. Compared with other reinforcement learning algorithms, the PPO algorithm has the characteristics of high sample efficiency and strong training stability, and is more suitable for complex scenarios with high dynamics and multiple constraints in loading operations.

[0076] like Figure 3 As shown, the intelligent decision-making model mainly includes a state encoder, a policy network, and a value network, with the specific design as follows:

[0077] (1) State encoder

[0078] The state encoder, serving as the input processing layer, includes a multimodal data fusion module and a temporal feature enhancement module. The multimodal data fusion module receives structured feature data such as material flow fluctuations and locomotive positioning deviations uploaded from edge computing nodes, as well as unstructured data such as visual images of material accumulation within the locomotive. It extracts spatial features from the images using a CNN, processes numerical features using a multilayer perceptron, and then dynamically assigns weights through an attention mechanism, outputting a unified state vector with a dimension of 128. The temporal feature enhancement module embeds an LSTM layer to perform temporal modeling of the state vectors from the past five time steps, capturing the dynamic correlation between "material drop speed changes and material accumulation trends," thus addressing the short-term decision-making bias caused by traditional encoders relying solely on single-frame data.

[0079] (2) Policy Network

[0080] The strategy network, serving as the action output layer, employs a two-layer architecture of "high-level strategy - low-level execution": the high-level strategy network generates macro-level strategies such as the material feeding allocation ratio and the division of the leveling machine's working area; the low-level execution network translates these macro-level strategies into specific control parameters, such as the opening degree of the material feeding gate, the joint angle of the leveling machine's robotic arm, and the fine-tuning amount of the machine's alignment. Furthermore, the constraint module incorporates hard constraint activation functions to ensure that the action space strictly conforms to safety thresholds, preventing the strategy network from outputting dangerous commands.

[0081] Traditional policy networks (PPOs) typically output the actions of a single agent, making it difficult to handle collaborative decision-making across multiple devices and prone to action conflicts. In contrast, the policy network output in this embodiment uses a device mask matrix to allocate independent output channels for the actions of different devices, ensuring logical consistency of actions across multiple devices.

[0082] (3) Value Network

[0083] The value network, serving as the evaluation feedback layer, extracts deep value features of the state through a shared feature extraction layer. It employs two parallel value network branches: an immediate value head to evaluate the impact of the current action on single loading efficiency and instantaneous energy consumption; and a long-term value head that combines historical operational data to evaluate the long-term impact of the strategy on equipment lifespan and system stability. Furthermore, a dynamic reward calibration module adjusts the evaluation weights of the value network in real-time based on deviations from the actual operational scenario, addressing the domain offset problem between the digital twin training environment and the physical world.

[0084] The dual objective function (minimizing energy consumption and maximizing efficiency) is the optimization objective of the intelligent decision-making model, expressed as:

[0085] maxJ = w1·J2 - w2·J1

[0086] min J1=∑(α1·P m +α2·P p +α3·Pl )

[0087] max J2=(M t / M target )×(T target / T operation )

[0088] In the formula: w1 and w2 are normalized weight coefficients, P m P p P l These are the locomotive traction power, the leveling machine drive power, and the unloading system management, respectively; α1, α2, and α3 are weighting coefficients; M t M represents the actual load capacity. target For the target load capacity, T target For the target operation time, T operation This refers to the actual working time.

[0089] The iterative training process is as follows: a training set containing several discrete scenarios is constructed in the digital twin environment. The scenario parameters include material characteristics, standard vehicle models, and typical working condition disturbances. A priority experience replay mechanism is adopted to oversample rare but high-risk scenarios such as equipment collisions and material spills. An attention mechanism is introduced to assign dynamic weights to key state features such as locomotive position deviation and carriage tilt. Adversarial examples are generated through self-adversarial training to improve the robustness of the model under abnormal working conditions.

[0090] Traditional PPO often favors one objective when dealing with two objectives due to fixed weights, and it is difficult to adapt to dynamic changes in the scenario. In contrast, this embodiment introduces an adaptive weight adjustment mechanism, which dynamically adjusts the weight coefficients of energy consumption and efficiency based on the current scenario parameters. At the same time, an objective conflict detector is designed in the value network, which triggers a preset compromise strategy when the two objectives are in extreme conflict.

[0091] A digital twin environment is a "virtual testing ground" for iterative model training. Its core is to reproduce the dynamic characteristics of real vehicle loading scenarios through physical mapping and parameter modeling. The specific construction steps are as follows:

[0092] (1) Digital mapping of physical entities: Through laser scanning, CAD drawing import and other methods, the three-dimensional dimensions and spatial position relationships of physical equipment such as loading towers, locomotives, and leveling machines are mapped to the virtual environment in a 1:1 ratio to form a high-precision geometric model and give the virtual equipment real physical parameters;

[0093] (2) Discretization and diversity design of scene parameters: Based on the principle of “covering typical scenes + extreme scenes”, the scene parameters are decomposed into three core dimensions (including: material characteristics, standard vehicle models, and typical working condition interference), and discrete scene samples are generated through orthogonal experiment method;

[0094] (3) Dynamic interactive logic modeling: Real-time simulation of the causal chain of "model decision-equipment action-scenario change". The core interactive logic includes: the dynamic relationship between the amount of material dropped and the stacking form of the carriage, the correlation between equipment action and energy consumption, and the dynamic evolution of the interference scenario.

[0095] Furthermore, the specific methods for generating the dynamic allocation strategy for material feeding and the timing planning of equipment actions are as follows:

[0096] Based on the current material characteristics, the location of the carriage, and the loading progress, the opening and closing timing and flow distribution coefficient of each material discharge port are output through the strategy network to form a material discharge scheme containing multiple control parameters.

[0097] The loading operation is decomposed into discrete control cycles of 100ms by using a time window sliding algorithm, generating a sequence of coordinated control instructions that includes locomotive travel speed curve, leveler insertion depth and timing, and material dropping system start and stop delay.

[0098] Furthermore, in the technical solution of this embodiment, the specific steps of S3 are as follows:

[0099] Edge computing nodes on fixed equipment build wired links through converged time-sensitive network switches, while edge computing nodes and autonomous decision-making modules on mobile devices access the network through 5G URLLC, solving the pain points of complex cabling and poor equipment mobility in traditional industrial networks.

[0100] Nanosecond-level clock alignment is performed on devices within the converged time-sensitive network to synchronize the 5G URLLC terminal time to the core network reference clock;

[0101] Define a standardized instruction format with timestamps. Edge computing nodes and autonomous decision-making modules interact with each other through a real-time communication protocol to provide device status feedback. Dynamically calibrate the material dropping start delay and the leveling machine action trigger point to ensure that the response time from the locomotive's arrival to the material dropping start is less than 50 milliseconds and the connection error from the end of material dropping to the leveling machine action is less than 20 milliseconds.

[0102] In this embodiment, the real-time communication module employs a real-time communication protocol that integrates Time-Sensitive Networking (TSN) and 5G URLLC technologies. TSN is a set of protocols under the IEEE 802.1 standard, designed specifically for scenarios such as industrial automation, achieving microsecond-level latency and nanosecond-level time synchronization through a deterministic communication mechanism. 5G URLLC is one of the three major 5G application scenarios defined by 3GPP, designed specifically for scenarios such as industrial control and telemedicine, aiming for air interface latency of less than 1 millisecond and reliability of 99.999%. Based on this, this embodiment can meet the stringent requirements of multi-device collaborative control and provide a standardized technical foundation for the expansion of subsequent scenarios.

[0103] Based on this, the real-time communication protocol can be improved to address the characteristics of loading operations, which include "high equipment mobility, complex interference sources, and close collaboration among multiple devices." Specifically, the protocol can be improved by: dynamically allocating TSN time windows and 5G slice resources according to service priority, achieving real-time binding of "priority-resource quota" to improve resource utilization; intelligently selecting "TSN main transmission + 5G backup," "5G main transmission + TSN backup," or single-link transmission modes based on channel quality and service importance to reduce unnecessary redundancy overhead and improve service reliability; and monitoring channel interference in real time and automatically adjusting modulation methods, transmission cycles, and retransmission strategies to improve transmission success rates in complex environments. The improved protocol is more suitable for the high real-time and high reliability requirements of multi-device collaboration in loading scenarios.

[0104] and Figure 1 Corresponding to the method described above, this embodiment of the invention also provides an edge computing-based autonomous decision-making and real-time communication system for loading depots, used for... Figure 1 The specific implementation of the method, the edge computing-based autonomous decision-making and real-time communication system for loading depots provided in this embodiment of the invention, can be applied to computer terminals or various mobile devices, such as... Figure 4 As shown, it specifically includes:

[0105] Edge computing node clusters are deployed in loading towers, unmanned locomotives, and intelligent leveling machines to form an "end-to-end" distributed processing architecture. They are used to collect real-time data and perform local processing, and upload key decision results to the autonomous decision-making module.

[0106] The autonomous decision-making module is used to construct a closed-loop mechanism of "state perception-strategy reasoning-instruction output" based on multi-source data, and autonomously generate material feeding control instructions and multi-device collaborative strategies.

[0107] The real-time communication module adopts a real-time communication protocol that integrates time-sensitive networking and 5G URLLC technology to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when devices execute commands.

[0108] In the intelligent loading system for railway transportation, the loading tower, unmanned locomotive, and intelligent leveling machine are the core equipment that work together to complete the material loading operation. Their working logic is as follows:

[0109] 1) Unmanned locomotives, as the transport carriers of materials, are responsible for automatically driving from the dispatch point to the designated loading position below the loading building, and transporting the fully loaded carriages away after the operation is completed;

[0110] 2) Loading tower: Based on the material characteristics and carriage parameters, the material feeding amount and feeding speed are dynamically controlled to accurately load the material into the carriage of the unmanned locomotive.

[0111] 3) Intelligent leveling machine: After the material is unloaded from the loading tower, it levels the unevenly piled material in the car to ensure that the material is evenly distributed and the height meets the transportation safety standards.

[0112] Through the division of labor of "locomotive positioning - loading tower loading - leveling machine sorting", the three achieve fully automated operation from empty carriages to fully loaded qualified carriages by combining intelligent decision-making and real-time communication, which greatly improves loading efficiency and quality, enhances the unmanned, highly collaborative and low-latency capabilities of loading operations, and thus better adapts to the needs of the "heavy-to-heavy" mode of railway collection and distribution.

[0113] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0114] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for autonomous decision-making and real-time communication in a loading terminal based on edge computing, characterized in that, Includes the following steps: S1. Deploy edge computing nodes at loading towers, unmanned locomotives, and intelligent leveling machines to build an "end-to-end" distributed processing architecture; Real-time data is collected through edge computing nodes and processed locally. The processed multi-source data is then uploaded to the autonomous decision-making module. S2. Utilizing the autonomous decision-making module, a closed-loop mechanism of "state perception - strategy reasoning - instruction output" is constructed based on the uploaded multi-source data to autonomously generate material dropping control instructions and multi-device collaborative strategies. S3. It adopts a real-time communication protocol that integrates Time-Sensitive Networking and 5G URLLC technology to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when equipment executes instructions, so as to realize the synchronization of locomotive arrival and material dropping start, and the connection between the leveling machine action and the end of material dropping. In S2, the autonomous decision-making module uses a proximal policy optimization algorithm to construct an intelligent decision-making model. The intelligent decision-making model includes a state encoder, a policy network, and a value network, as detailed below: The state encoder serves as the input processing layer, including a multimodal data fusion module and a temporal feature enhancement module; The policy network, as the action output layer, adopts a two-layer architecture of "high-level policy - low-level execution". The high-level policy network generates macro policies, and the low-level execution network converts the macro policies into specific control parameters. The output end of the policy network uses a device mask matrix to allocate independent output channels for the actions of different devices, ensuring the logical consistency of actions of multiple devices. The value network, as the evaluation feedback layer, extracts deep value features from the multi-dimensional state vector output by the state encoder through a shared feature extraction layer. The value network introduces an adaptive weight adjustment mechanism to dynamically adjust the weight coefficients of energy consumption and efficiency based on the current scenario parameters. At the same time, a target conflict detector is designed in the value network to trigger a preset compromise strategy when two targets are in extreme conflict. The specific steps for S3 are as follows: Edge computing nodes of fixed equipment build wired links through converged time-sensitive network switches, while edge computing nodes and autonomous decision-making modules of mobile devices access the network through 5G URLLC. Nanosecond-level clock alignment is performed on devices within the converged time-sensitive network to synchronize the 5G URLLC terminal time to the core network reference clock; The real-time communication module defines a standardized instruction format with timestamps. Edge computing nodes and autonomous decision-making modules interact with each other through the real-time communication protocol to exchange device status and execution feedback. This dynamically calibrates the material dropping start delay and the leveling machine action trigger point, ensuring that the response time from the locomotive's arrival to the material dropping start is less than 50 milliseconds and the connection error from the end of material dropping to the leveling machine action is less than 20 milliseconds.

2. The method for autonomous decision-making and real-time communication of loading depots based on edge computing according to claim 1, characterized in that, In S1, an "end-to-end" distributed processing architecture is constructed, specifically as follows: The loading tower is equipped with a first edge computing node with multi-interface data acquisition function. The first edge computing node is connected to the loading tower's material unloading control unit, material level sensor, and weighing device through an industrial bus and is deployed in the loading tower's control cabinet. The unmanned locomotive is equipped with a second edge computing node that has GNSS positioning and vehicle bus adaptation functions. The second edge computing node is connected to the locomotive's automatic driving system and carriage positioning sensors via vehicle Ethernet and is deployed in the electrical cabinet of the unmanned locomotive. A third edge computing node with motion control signal processing function is configured for the intelligent leveling machine. The third edge computing node is connected to the leveling machine's actuator and attitude sensor through a wireless communication module and is deployed in the protective cabin of the intelligent leveling machine. Physical connections between edge computing nodes are established through a fiber optic ring network, and a distributed consensus algorithm is used to achieve clock synchronization and data sharding storage between nodes, forming an end-to-end redundant communication link.

3. The method for autonomous decision-making and real-time communication of loading depots based on edge computing according to claim 1, characterized in that, In S1, real-time data is collected and processed locally, specifically as follows: Edge computing nodes are used to collect data on material flow rate, material level in silos, and gate opening of the loading tower; real-time location, speed, carriage number and size data of unmanned locomotives; and working position, actuator posture and drive motor current data of intelligent leveling machines. The collected real-time data is filtered, denoised, timestamped, and normalized. Key features are extracted, and the abnormal state of the equipment is judged based on a preset threshold. Structured data containing the equipment operating status identifier is generated and uploaded to the autonomous decision-making module.

4. The method for autonomous decision-making and real-time communication of loading depots based on edge computing according to claim 1, characterized in that, The specific steps of S2 are as follows: Spatiotemporal fusion of multi-source data is performed, and parameters related to equipment operating status and working environment are aggregated through feature weighting algorithms to construct a multi-dimensional state vector; A deep reinforcement learning algorithm is adopted, with the dual objective functions of minimizing energy consumption and maximizing loading efficiency. Iterative training is carried out in a variety of preset scenarios to generate a dynamic material loading allocation strategy and equipment action timing plan. The dynamic material feeding allocation strategy is converted into standardized control commands, which are then sent to the corresponding devices via the S3 real-time communication protocol. The system also receives device execution feedback data and updates the state vector to complete the closed-loop iteration.

5. The method for autonomous decision-making and real-time communication of loading depots based on edge computing according to claim 4, characterized in that, The specific methods for generating dynamic material feeding allocation strategies and equipment action timing planning are as follows: Based on the current material characteristics, the location of the carriage, and the loading progress, the opening and closing timing and flow distribution coefficient of each material discharge port are output through the strategy network to form a material discharge scheme containing multiple control parameters. The loading operation is decomposed into discrete control cycles of 100ms by using a time window sliding algorithm, generating a sequence of coordinated control instructions that includes locomotive travel speed curve, leveler insertion depth and timing, and material dropping system start and stop delay.

6. A system for implementing the edge computing-based autonomous decision-making and real-time communication method for loading depots as described in any one of claims 1-5, characterized in that, include: Edge computing node clusters are deployed in loading towers, unmanned locomotives, and intelligent leveling machines to form an "end-to-end" distributed processing architecture. They are used to collect real-time data and perform local processing, and upload key decision results to the autonomous decision-making module. The autonomous decision-making module is used to construct a closed-loop mechanism of "state perception-strategy reasoning-instruction output" based on multi-source data, and autonomously generate material dropping control instructions and multi-device collaborative strategies. The real-time communication module adopts a real-time communication protocol that integrates time-sensitive networking and 5G URLLC technology to ensure real-time communication between edge computing nodes, between edge computing nodes and autonomous decision-making modules, and when devices execute commands.

Images