A method, device, and electronic equipment for dynamic adjustment of traction control of heavy-haul trains.

Abstract

This invention provides a method, device, and electronic equipment for dynamic adjustment of traction control for heavy-haul trains. The method includes: real-time acquisition of the operating status, track environment, and coupler force data of the heavy-haul train to obtain status data; using a pre-built communication network as a carrier, generating decision commands for dynamic traction force allocation based on the status data and a multi-objective optimization strategy; converting the decision commands into corresponding physical actions according to the locomotive model of the heavy-haul train; collecting real-time operating data of the heavy-haul train, and updating the decision commands based on the real-time operating data. This invention achieves the wireless multiple-unit synchronous control performance requirements under different operating conditions, communication delays, and mixed multiple-unit operation of different locomotive models, solving the problem of poor traction force adjustment performance of heavy-haul trains in existing related technologies, and ensuring the safe operation of heavy-haul combined trains.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for heavy-haul railways, and in particular to a dynamic adjustment method, device, and electronic equipment for traction control of heavy-haul trains. Background Technology

[0002] Currently, there is limited research on dynamic traction adjustment methods in the heavy-haul train industry. Chinese invention patent CN114633780A proposes a heavy-haul train and its longitudinal dynamic traction operation optimization control system. This system utilizes model prediction to suppress longitudinal impulses during gradient change point operating condition switching, and optimizes train dynamic performance by dynamically adjusting the traction and braking force differences between the master and slave locomotives. Chinese invention patent CN113942544A proposes a wireless multiple-unit remote distributed power traction operation control system and multiple-unit locomotives. By integrating train control and management systems, braking control units, etc., it achieves asynchronous control of mixed locomotives, gradually transitioning to power coordination under different operating conditions. Both patents propose dynamically adjusting traction power through real-time data interaction between the master and slave locomotives, reducing the risk of control failure due to communication interruptions.

[0003] Existing model predictive control still exhibits lag in response to abrupt changes in track parameters (such as gradient switching and curve resistance variations), especially in long-formation trains, where the peak coupler force suppression effect is limited. Significant differences in traction response characteristics and braking delays among different locomotive models lead to decreased synchronization accuracy in mixed-formation scenarios. Although differentiated control units have partially addressed this issue, the compatibility of existing algorithms with nonlinear characteristics still needs optimization. Wireless communication is susceptible to interference in complex terrain (tunnels, mountainous areas). While dual-channel redundancy reduces risk, command packet loss or delays (>100ms) may still occur in extreme environments, affecting the real-time performance of traction force adjustment. Dynamic traction force adjustment may increase energy consumption, especially in scenarios with frequent speed adjustments. Existing research primarily focuses on safety, with less comprehensive consideration of energy efficiency optimization.

[0004] There is currently no effective solution to the problem of poor traction adjustment performance of heavy-haul trains in existing related technologies. Summary of the Invention

[0005] This invention provides a dynamic adjustment method, device, and electronic equipment for traction control of heavy-haul trains, in order to solve the defects of poor traction force adjustment performance of heavy-haul trains in existing related technologies.

[0006] In a first aspect, the present invention provides a dynamic adjustment method for traction control of heavy-haul trains, comprising:

[0007] Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data is used to obtain status data.

[0008] Using a pre-built communication network as a carrier, and based on the state data, a decision command for dynamic allocation of traction force is generated according to a multi-objective optimization strategy.

[0009] Based on the locomotive model of the heavy-haul train, the decision command is converted into a corresponding physical action;

[0010] Collect real-time operating data of the heavy-haul train and update the decision-making instructions based on the real-time operating data.

[0011] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, which collects real-time data on the operating status of the heavy-haul train, the track environment, and the coupler force data to obtain status data, including:

[0012] The operating status and coupler force data of the heavy-haul train are collected through a multi-source sensor network;

[0013] The route's gradient, radius of curvature, and tunnel locations are obtained from a digital map of the route.

[0014] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, which collects the operating status and coupler force data of the heavy-haul train through a multi-source sensor network, including:

[0015] The longitudinal tension / compression of the heavy-haul train is collected by the coupler force sensor;

[0016] The locomotive acceleration and attitude angle of the heavy-haul train are detected by the inertial navigation unit;

[0017] The temperature of the wheelset bearings is monitored in real time using an infrared thermal imager.

[0018] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, wherein the coupler force sensor is deployed in the coupler buffer device;

[0019] The inertial navigation unit is used for slope compensation calculation.

[0020] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the communication network includes a dual-mode redundant communication network and edge computing nodes;

[0021] The dual-mode redundant communication network includes a dedicated 5G-R network and a low-Earth orbit satellite link.

[0022] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the 5G-R dedicated network is used for conventional command transmission, the network bandwidth is above the bandwidth threshold, the network latency is within the first latency threshold, and multi-locomotive multicast communication is supported.

[0023] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, wherein the low-orbit satellite link is an emergency communication channel with a channel delay within a second delay threshold, and is deployed on each locomotive of the heavy-haul train.

[0024] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the optimized communication protocol of the communication network includes dynamic priority scheduling and a breakpoint resume mechanism.

[0025] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein in the dynamic priority scheduling, data packets are classified and a differentiated QoS strategy is adopted.

[0026] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, wherein the data packet includes S0 emergency braking command, S1 traction adjustment command and S2 status monitoring data.

[0027] According to the present invention, the traction control dynamic adjustment method for heavy-haul trains is provided in which the S0 emergency braking command is transmitted via satellite link, the delay is within the second delay threshold, and the number of data packet retransmissions is more than 3.

[0028] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the S1 traction adjustment command is transmitted through the 5G-R dedicated network with a delay within a first delay threshold and forward error correction coding is used.

[0029] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, which generates a decision command for dynamic traction force allocation based on a multi-objective optimization strategy according to the state data, including:

[0030] The digital twin model parameters are updated based on the state data, and a basic control sequence is generated using a model predictive control algorithm.

[0031] The Actor-Critic network is trained using a deep reinforcement learning algorithm to learn the optimal strategy under complex working conditions.

[0032] The strategy risk is assessed through Monte Carlo tree search to determine the decision instructions for dynamic traction allocation.

[0033] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the objective function of the model predictive control algorithm is:

[0034]

[0035] in, J Let N represent the objective function, N represent the prediction time domain, and k represent the time step index. , These are the time-varying weighting factors for the coupler force term and energy consumption term, respectively. This represents the coupler force at the k-th predicted time. This represents the traction power or energy consumption at the k-th prediction time.

[0036] According to the present invention, a dynamic adjustment method for traction control of heavy-haul trains is provided, wherein the edge computing node includes:

[0037] The communication quality perception module monitors the channel quality index in real time and triggers communication mode switching.

[0038] The computational task offloading strategy dynamically distributes Monte Carlo tree inference tasks to cloud and edge nodes.

[0039] According to the dynamic adjustment method for traction control of heavy-haul trains provided by the present invention, the rule for the communication quality sensing module to trigger communication mode switching is as follows:

[0040]

[0041] The formula for assigning weights to the computational task unloading strategy is as follows:

[0042]

[0043] in, W edge This represents the dynamic weighting coefficient.

[0044] According to the dynamic adjustment method for traction control of heavy-haul trains provided by the present invention, when the decision command is converted into the corresponding physical action, it is necessary to satisfy the dynamic relaxation transition strategy and the safety limiting mechanism.

[0045] The dynamic relaxation transition strategy is as follows: when the gradient changes abruptly, the main locomotive of the heavy-haul train outputs traction force in advance, and the slave locomotive follows according to an exponential curve.

[0046] The safety limiting mechanism is as follows: real-time monitoring of the actuator status; if the output deviates from the decision command by more than a preset standard, self-check and degraded operation are triggered.

[0047] According to the present invention, a dynamic adjustment method for traction control of a heavy-haul train is provided, which collects real-time operating data of the heavy-haul train and updates the decision command based on the real-time operating data, including:

[0048] Collect real-time operating data of the heavy-haul train and upload it to the cloud digital twin;

[0049] The Monte Carlo tree is trained offline to generate new decision instructions;

[0050] A / B testing is conducted on both the new and old decision instructions, and the best decision instruction is updated based on the test results.

[0051] Secondly, the present invention also provides a dynamic adjustment device for traction control of heavy-haul trains, comprising:

[0052] The sensing module is used to collect real-time data on the operating status of heavy-haul trains, track environment, and coupler stress to obtain status data.

[0053] The communication module is used to ensure real-time interaction between control commands and status data, and to build a redundant communication network;

[0054] The decision module is used to generate decision instructions for dynamic allocation of traction force based on the state data and a multi-objective optimization strategy.

[0055] The execution module is used to convert the decision command into a corresponding physical action based on the locomotive model of the heavy-haul train.

[0056] An optimization module is used to collect real-time operating data of the heavy-haul train and update the decision instructions based on the real-time operating data.

[0057] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the dynamic adjustment method for traction control of heavy-haul trains as described in the first aspect above.

[0058] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the dynamic adjustment method for traction control of heavy-haul trains as described in the first aspect above.

[0059] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the dynamic adjustment method for traction control of heavy-haul trains as described in the first aspect above.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The dynamic adjustment method for traction control of heavy-haul trains provided by this invention can effectively avoid the lag in response to sudden changes in track parameters (such as gradient switching and changes in curve resistance), especially in long-formation trains where the peak coupler force suppression effect is limited. Furthermore, this method offers better compatibility with nonlinear characteristics and better real-time traction force adjustment. This method achieves the wireless synchronous control performance requirements for multiple-unit trains operating under different conditions, communication delays, and mixed formations of different locomotive models, solving the problem of poor traction force adjustment performance in existing related technologies and ensuring the safe operation of heavy-haul combined trains. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0063] Figure 1 This is a flowchart of the dynamic adjustment method for traction control of heavy-haul trains provided by the present invention;

[0064] Figure 2 This is a structural block diagram of the dynamic adjustment device for traction control of heavy-haul trains provided by the present invention.

[0065] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0067] This invention provides a dynamic adjustment method for traction control of heavy-haul trains. Figure 1 This is a flowchart of the dynamic adjustment method for traction control of heavy-haul trains provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0068] Step S101: Real-time collection of the operating status of heavy-haul trains, track environment, and coupler force data to obtain status data;

[0069] Step S102: Using a pre-built communication network as a carrier, a decision command for dynamic allocation of traction force is generated based on the state data and a multi-objective optimization strategy.

[0070] Step S103: Based on the locomotive model of the heavy-haul train, the decision command is converted into the corresponding physical action;

[0071] Step S104: Collect real-time operation data of heavy-haul trains and update decision instructions based on the real-time operation data.

[0072] In this method, firstly, real-time data on the operating status of heavy-haul trains, track environment, and coupler force are collected to provide input for control decisions. Then, the status data is transmitted via a pre-built communication network. Based on the status data, a decision command for dynamic traction force allocation is generated using a multi-objective optimization strategy to coordinate the actions of heterogeneous locomotives. The decision command is then converted into physical actions, adapting to the actuators of different locomotive models for precise dynamic response and compatible control. Finally, data-driven optimization of model parameters and decision commands improves system adaptability. This process effectively avoids lag in response to sudden changes in track parameters (such as gradient changes and curve resistance variations), especially in long-formation trains where the peak coupler force suppression effect is limited. Furthermore, this method offers better compatibility with nonlinear characteristics and better real-time traction force adjustment. This method achieves the wireless multiple-unit synchronous control performance requirements for different operating conditions, communication delays, and mixed multiple-unit trains of different models, solving the problem of poor traction force adjustment performance in existing related technologies and ensuring the safe operation of heavy-haul combined trains.

[0073] Furthermore, the above steps are executed sequentially as a positive control flow. The above steps of the present invention also provide an execution strategy for a reverse feedback flow, namely the execution order of step 104 → step S101 → step S102 → step S103, which realizes a closed-loop data flow and feedback mechanism.

[0074] In some embodiments, step S101 involves real-time acquisition of the operating status of the heavy-haul train, the track environment, and the coupler stress data to obtain status data, including: acquiring the operating status of the heavy-haul train and the coupler stress data through a multi-source sensor network; and obtaining pre-stored track gradient, radius of curvature, and tunnel location through a digital map of the track.

[0075] In this embodiment, the operating status and coupler force data of the heavy-haul train are collected through a multi-source sensor network, including: collecting the longitudinal tension / compression of the heavy-haul train through a coupler force sensor; detecting the locomotive acceleration and attitude angle of the heavy-haul train through an inertial navigation unit; and monitoring the wheel bearing temperature in real time through an infrared thermal imager.

[0076] For example, a coupler force sensor is deployed in the coupler buffer device to measure longitudinal tension / compression (accuracy ±1kN, sampling frequency 100Hz); an inertial measurement unit (IMU) monitors locomotive acceleration and attitude angles (lateral / longitudinal tilt) for gradient compensation calculations; axle temperature monitoring uses an infrared thermal imager to detect wheelset bearing temperature in real time to prevent sudden changes in adhesion coefficient. The digital map of the route pre-stores geographical information such as route gradient, radius of curvature, and tunnel locations, supporting road condition prediction up to 3km ahead.

[0077] In some of these embodiments, the communication network includes a dual-mode redundant communication network and edge computing nodes; the dual-mode redundant communication network includes a 5G-R dedicated network and a low-Earth orbit satellite link.

[0078] Specifically, this method employs a dual-mode redundant communication network (5G-R + satellite link) + edge computing nodes (local decision-making to reduce latency). The dual-mode communication channels are as follows: a dedicated 5G-R network is used for regular command transmission, with network bandwidth exceeding the bandwidth threshold and network latency within the first latency threshold (e.g., bandwidth ≥ 100Mbps, latency ≤ 50ms), supporting multicast communication between multiple locomotives; a low-Earth orbit satellite link (e.g., Starlink) serves as an emergency communication channel, with channel latency within the second latency threshold (e.g., latency ≤ 200ms), ensuring coverage in blind areas such as mountainous regions and tunnels. Edge computing nodes are deployed on each locomotive to achieve localized model inference and control decision caching, reducing cloud dependence and mitigating the impact of latency on decision-making.

[0079] In this embodiment, optimizing the communication protocol includes:

[0080] Dynamic priority scheduling: Data packets are divided into three levels (S0 emergency braking command, S1 traction adjustment command, and S2 status monitoring data), and a differentiated QoS strategy is adopted. Among them, S0 level commands are transmitted through satellite links, with latency within the second latency threshold, for example, latency ≤200ms, and data packet retransmission count ≥3; S1 level commands are transmitted through 5G-R, with latency within the first latency threshold, for example, latency ≤50ms, and forward error correction (FEC) coding is adopted.

[0081] Resume interruption mechanism: Data packets are sorted and resent based on timestamps to ensure data integrity after communication is interrupted.

[0082] Based on this, in step S102, a decision instruction for dynamic traction allocation is generated according to the state data and a multi-objective optimization strategy. This includes: updating the parameters of the digital twin model according to the state data; generating a basic control sequence through a model predictive control (MPC) algorithm; training an Actor-Critic network through a deep reinforcement learning (DRL) algorithm to learn the optimal strategy under complex working conditions; and evaluating the strategy risk through Monte Carlo Tree Search (MCTS) to determine the decision instruction for dynamic traction allocation.

[0083] Furthermore, the edge computing node includes: a communication quality awareness module, which monitors the channel quality index in real time and triggers communication mode switching; and a computing task offloading strategy, which dynamically allocates Monte Carlo tree inference tasks to the cloud and edge nodes.

[0084] In this embodiment, a cluster of dynamic optimization algorithms based on digital twins is employed, including model prediction control algorithms, deep reinforcement learning algorithms, and Monte Carlo tree search evaluation strategies.

[0085] Real-time reception of data from the perception layer updates the parameters of the digital twin model. A basic control sequence is generated using a model predictive control algorithm (MMC). The MMC has a rolling optimization window of 10 seconds, and the objective function is:

[0086]

[0087] in, J Let N represent the objective function, N represent the prediction time domain (representing the total number of future time steps the controller predicts forward. For example, if each time step is 1 second, N=10 means the controller considers the situation 10 seconds ahead in each optimization), and k represent the time step index (representing the k-th time in the prediction time domain, k ranges from 1 to N). , These are the time-varying weighting factors for the coupler force term and energy consumption term, respectively. This represents the coupler force at the k-th predicted time. This represents the traction power or energy consumption at the k-th prediction time.

[0088] Deep reinforcement learning algorithms train Actor-Critic networks in digital twin platforms to learn optimal strategies under complex conditions (such as continuous ramps and mixed-track operations), providing corrections for decision-making.

[0089] Finally, the strategy risk is assessed through Monte Carlo tree search, and the final decision instruction is output. For the cooperative strategy of heterogeneous locomotives, a feedforward-feedback composite controller is designed based on the characteristic database (e.g., the traction response delay of HXD locomotive is 0.8s, and that of Dongfeng 4 locomotive is 1.2s) to compensate for dynamic differences.

[0090] Edge computing nodes perform local MPC calculations, and the cloud-based DRL model pushes updated parameters every 10 minutes. In the event of a communication interruption, the edge node makes an autonomous decision based on the last valid instruction, with a maximum sustained control period of 10 seconds. The rules for the communication quality awareness module to trigger communication mode switching are as follows:

[0091]

[0092] The formula for calculating the weights assigned to the task unloading strategy is as follows:

[0093]

[0094] in, W edge This represents the dynamic weighting coefficient (its value is between 0 and 1). It represents the degree or proportion to which a computational task (specifically the inference task of the DRL model in this scheme) is assigned to edge computing nodes (i.e., on-board computing devices) for execution.

[0095] For example, a hybrid algorithm combining model predictive control and deep reinforcement learning at the decision layer includes:

[0096] 1. Rolling Optimization Module: This module is used to calculate the traction force distribution sequence in a preset time window based on the train's longitudinal dynamics model. The objective function is to minimize the weighted sum of the peak coupler force and energy consumption.

[0097] 2. DRL Compensation Module: Generates traction correction through a deep reinforcement learning network, with inputs including real-time coupler force, track gradient, and locomotive formation status;

[0098] 3. Risk Assessment Module: Based on Monte Carlo tree search, this module assesses the risk of the combined strategy of MPC basic instructions and DRL corrections, and outputs the final control instructions.

[0099] 4. Hardware acceleration unit: The objective function is solved in parallel using a field programmable gate array (FPGA), which includes multiple parallel computing units and a dynamic task scheduler.

[0100] In some embodiments, step S103, when converting the decision command into the corresponding physical action, needs to satisfy a dynamic relaxation transition strategy and a safety limiting mechanism. The dynamic relaxation transition strategy is: when there is a sudden change in gradient, the main control locomotive of the heavy-haul train outputs 5% traction force in advance, and the slave control locomotive follows according to an exponential curve. The safety limiting mechanism is: the actuator status is monitored in real time, and if the output deviates from the decision command by more than a preset standard, self-check and degraded operation are triggered. Preferably, the preset standard is 10%.

[0101] In this embodiment, a differentiated control unit (compatible with heterogeneous locomotives) and adaptive actuators (electronic proportional valve and traction converter) are used for command conversion and action execution. The differentiated control unit supports multiple control protocols (such as Train Control and Management System (TCMS) and Locotrol), achieving heterogeneous system compatibility through virtualization middleware. The adaptive actuators include an electronic proportional valve and a traction converter. The electronic proportional valve has a response time of less than 20ms and a traction force adjustment accuracy of ±2kN; the traction converter uses Model-Referencing Adaptive Control (MRAC) to dynamically adjust the Pulse Width Modulation (PWM) frequency to match load changes.

[0102] Based on the above embodiments, step S104, collecting real-time operation data of heavy-haul trains and updating decision instructions based on real-time operation data, includes: collecting real-time operation data of heavy-haul trains and uploading it to a cloud-based digital twin; training a Monte Carlo tree offline to generate new decision instructions; conducting A / B testing on the new and old decision instructions, and updating the decision instructions based on the best test results.

[0103] In this embodiment, a parameter self-learning module (online parameter identification) and a digital twin simulation platform (strategy iteration) are used to analyze data and optimize decision commands. The online parameter identification process is as follows: recursive least squares with a forgetting factor (FFRLS) is used to update the coupler stiffness and damping coefficient every 30 seconds. For the digital twin simulation platform, a high-fidelity virtual train (1:1 physical mapping) is constructed, historical operating data (such as the 10-year operating condition database of the Daqin Railway) is injected, and the DRL strategy is trained. Multi-objective energy efficiency optimization is achieved through the above modules. Specifically, based on the NSGA-II genetic algorithm, the Pareto optimal solution is found under safety constraints to balance coupler force suppression and energy consumption reduction. The optimization process is as follows: real-time operating data is uploaded to the cloud digital twin; offline batch training of the DRL strategy generates new control rules; updates are wirelessly pushed to the on-board edge node via over-the-air (OTA) technology; A / B testing of the old and new strategies is conducted, and the optimal strategy is deployed.

[0104] In summary, this method adopts a five-layer closed-loop architecture of "perception-communication-decision-execution-optimization", which supports multi-locomotive mixed formation and dynamic adaptation to complex working conditions. Through layer-by-layer closed loops and dynamic feedback, this architecture systematically solves the problems of synchronization, safety and energy efficiency in the dynamic adjustment of traction force of heavy-haul trains, and provides a reliable technical foundation for heavy-haul transportation under complex working conditions.

[0105] The present invention also provides a dynamic adjustment device for traction control of heavy-haul trains. The dynamic adjustment device for traction control of heavy-haul trains provided by the present invention will be described below. The dynamic adjustment device for traction control of heavy-haul trains described below can be referred to in correspondence with the dynamic adjustment method for traction control of heavy-haul trains described above. Figure 2 This is a structural block diagram of the dynamic adjustment device for traction control of heavy-haul trains provided by the present invention, as shown below. Figure 2 As shown, the device includes:

[0106] The sensing module 201 is used to collect real-time data on the operating status of heavy-haul trains, track environment, and coupler force to obtain status data.

[0107] Communication module 202 is used to ensure real-time interaction between control commands and status data and to build a redundant communication network;

[0108] The decision module 203 is used to generate decision instructions for dynamic allocation of traction force based on the status data and a multi-objective optimization strategy.

[0109] The execution module 204 is used to convert decision commands into corresponding physical actions based on the locomotive model of the heavy-haul train.

[0110] The optimization module 205 is used to collect real-time operating data of heavy-haul trains and update decision instructions based on the real-time operating data.

[0111] In operation, this device first collects real-time data on the operating status of heavy-haul trains, track environment, and coupler force, providing input for control decisions. Then, the communication module 202 transmits the status data via a pre-built communication network. Based on the status data, the decision module 203 generates dynamic traction force allocation instructions using a multi-objective optimization strategy, coordinating the actions of different locomotives. The execution module 204 then translates the decision instructions into physical actions, adapting to different locomotive actuators for precise dynamic response and compatible control. Finally, the optimization module 205 optimizes model parameters and decision instructions through data-driven optimization, improving system adaptability. This process effectively avoids lag in response to sudden changes in track parameters (such as gradient changes and curve resistance variations), especially in long-formation trains where coupler force peak suppression is limited. Furthermore, this device offers better compatibility with nonlinear characteristics and better real-time traction force adjustment. This device achieves the wireless multiple-unit synchronous control performance requirements for different operating conditions, communication delays, and mixed multiple-unit operation of different locomotive models. It solves the problem of poor traction adjustment performance of heavy-haul trains in existing related technologies and ensures the safe operation of heavy-haul combined trains.

[0112] Specifically, the perception module 201 includes a coupler force sensor, an inertial navigation unit, axle temperature detection, a digital map of the track, and existing data. It can fuse data from multiple sensors to eliminate noise interference from a single sensor. Furthermore, it employs a deep learning model to predict short-term coupler force changes and trigger adjustments to the control strategy in advance.

[0113] The communication module 202 includes a 5G-R dedicated network, a low-orbit satellite link, and an edge computing node. It can divide data packets into three levels, adopt differentiated QoS strategies, and sort and resend data packets based on timestamps to ensure the integrity of data after the communication terminal.

[0114] The decision module 203 integrates a model predictive control algorithm, a deep reinforcement learning algorithm, and a Monte Carlo tree evaluation algorithm. It can receive data from the perception layer in real time and update the parameters of the digital twin model. Specifically, the model predictive control algorithm generates the basic control sequence, the deep reinforcement learning algorithm provides correction parameters, and finally, the Monte Carlo tree search evaluates the policy risk and outputs the final instruction.

[0115] The execution module 204 includes a differentiated control unit and an adaptive actuator. When there is a sudden change in gradient, the main control locomotive outputs traction force 5% in advance, and the slave control locomotive follows according to an exponential curve to avoid step impact. The actuator status is monitored, and if the output deviates from the command by more than 10%, self-test and degraded operation are triggered.

[0116] The optimization module 205 includes a parameter self-learning module and a digital twin simulation platform. It can upload real-time operational data to the cloud-based digital twin platform, conduct offline batch training of deep reinforcement learning algorithms to generate new control rules, and wirelessly push updates to the vehicle's edge nodes via OTA. Furthermore, it performs A / B testing on the old and new strategies and deploys the optimal one.

[0117] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 301, a communication interface 302, a memory 303, and a communication bus 304. The processor 301, communication interface 302, and memory 303 communicate with each other via the communication bus 304. The processor 301 can call logical instructions from the memory 303 to execute a dynamic adjustment method for traction control of heavy-haul trains. This method includes:

[0118] Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data is used to obtain status data.

[0119] Using a pre-built communication network as a carrier, and based on state data, decision instructions for dynamic allocation of traction force are generated according to a multi-objective optimization strategy.

[0120] Based on the locomotive model of the heavy-haul train, the decision-making instructions are translated into corresponding physical actions;

[0121] Collect real-time operation data of heavy-haul trains and update decision-making instructions based on the real-time operation data.

[0122] Furthermore, the logical instructions in the aforementioned memory 303 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the dynamic adjustment method for traction control of heavy-haul trains provided by the above methods, the method comprising:

[0124] Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data is used to obtain status data.

[0125] Using a pre-built communication network as a carrier, and based on state data, decision instructions for dynamic allocation of traction force are generated according to a multi-objective optimization strategy.

[0126] Based on the locomotive model of the heavy-haul train, the decision-making instructions are translated into corresponding physical actions;

[0127] Collect real-time operation data of heavy-haul trains and update decision-making instructions based on the real-time operation data.

[0128] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the dynamic adjustment method for traction control of heavy-haul trains provided by the methods described above, the method comprising:

[0129] Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data is used to obtain status data.

[0130] Using a pre-built communication network as a carrier, and based on state data, decision instructions for dynamic allocation of traction force are generated according to a multi-objective optimization strategy.

[0131] Based on the locomotive model of the heavy-haul train, the decision-making instructions are translated into corresponding physical actions;

[0132] Collect real-time operation data of heavy-haul trains and update decision-making instructions based on the real-time operation data.

[0133] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0134] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic adjustment method for traction control of heavy-haul trains, characterized in that, include: Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data is used to obtain status data. Using a pre-built communication network as a carrier, a decision command for dynamic traction allocation is generated based on the state data and a multi-objective optimization strategy; the communication network includes edge computing nodes, which are deployed on each locomotive. Based on the locomotive model of the heavy-haul train, the decision command is converted into a corresponding physical action; Based on the state data, a decision instruction for dynamic traction allocation is generated using a multi-objective optimization strategy, including: The digital twin model parameters are updated based on the state data, and a basic control sequence is generated using a model predictive control algorithm. The Actor-Critic network is trained using a deep reinforcement learning algorithm to learn the optimal strategy under complex working conditions. The strategy risk is assessed through Monte Carlo tree search to determine the decision instructions for dynamic traction allocation; The real-time operation data of the heavy-haul train is collected and uploaded to the cloud digital twin. The deep reinforcement learning algorithm strategy is trained offline in batches to generate new control rules.

2. The dynamic adjustment method for traction control of heavy-haul trains according to claim 1, characterized in that, Real-time data collection of the operating status of heavy-haul trains, track environment, and coupler stress data yields status data, including: The operating status and coupler force data of the heavy-haul train are collected through a multi-source sensor network; The route's gradient, radius of curvature, and tunnel locations are obtained from a digital map of the route.

3. The dynamic adjustment method for traction control of heavy-haul trains according to claim 2, characterized in that, The operating status and coupler force data of the heavy-haul train are collected through a multi-source sensor network, including: The longitudinal tension / compression of the heavy-haul train is collected by the coupler force sensor; The locomotive acceleration and attitude angle of the heavy-haul train are detected by the inertial navigation unit; The temperature of the wheelset bearings is monitored in real time using an infrared thermal imager.

4. The dynamic adjustment method for traction control of heavy-haul trains according to claim 3, characterized in that, The coupler force sensor is deployed in the coupler buffer device; The inertial navigation unit is used for slope compensation calculation.

5. The dynamic adjustment method for traction control of heavy-haul trains according to claim 1, characterized in that, The communication network includes a dual-mode redundant communication network; The dual-mode redundant communication network includes a dedicated 5G-R network and a low-Earth orbit satellite link.

6. The dynamic adjustment method for traction control of heavy-haul trains according to claim 5, characterized in that, The 5G-R dedicated network is used for regular command transmission, with network bandwidth exceeding the bandwidth threshold, network latency within the first latency threshold, and supports multi-vehicle multicast communication.

7. The dynamic adjustment method for traction control of heavy-haul trains according to claim 5, characterized in that, The low-orbit satellite link is an emergency communication channel with a channel delay within a second delay threshold, and is deployed on each locomotive of the heavy-haul train.

8. The dynamic adjustment method for traction control of heavy-haul trains according to claim 5, characterized in that, The optimized communication protocol of the communication network includes dynamic priority scheduling and a breakpoint resume mechanism.

9. The dynamic adjustment method for traction control of heavy-haul trains according to claim 8, characterized in that, In the dynamic priority scheduling, data packets are classified and a differentiated QoS strategy is adopted.

10. The dynamic adjustment method for traction control of heavy-haul trains according to claim 9, characterized in that, The data packets are categorized into S0 emergency braking commands, S1 traction adjustment commands, and S2 status monitoring data.

11. The dynamic adjustment method for traction control of heavy-haul trains according to claim 9, characterized in that, The S0 emergency braking command is transmitted via satellite link, with a delay within the second delay threshold and a data packet retransmission count of more than 3 times.

12. The dynamic adjustment method for traction control of heavy-haul trains according to claim 9, characterized in that, The S1 traction adjustment command is transmitted through the 5G-R dedicated network with a latency within the first latency threshold and uses forward error correction coding.

13. The dynamic adjustment method for traction control of heavy-haul trains according to claim 5, characterized in that, The objective function of the model predictive control algorithm is: in, J Let N represent the objective function, N represent the prediction time domain, and k represent the time step index. , These are the time-varying weighting factors for the coupler force term and energy consumption term, respectively. This represents the coupler force at the k-th predicted time. This represents the traction power or energy consumption at the k-th prediction time.

14. The dynamic adjustment method for traction control of heavy-haul trains according to claim 5, characterized in that, The edge computing nodes include: The communication quality perception module monitors the channel quality index in real time and triggers communication mode switching. The computational task offloading strategy dynamically distributes Monte Carlo tree inference tasks to cloud and edge computing nodes.

15. The dynamic adjustment method for traction control of heavy-haul trains according to claim 14, characterized in that, The rule for the communication quality sensing module to trigger communication mode switching is as follows: The formula for assigning weights to the computational task unloading strategy is as follows: in, W edge This represents the dynamic weighting coefficient, indicating the degree or proportion to which a computational task is recommended to be assigned to an edge computing node for execution.

16. The dynamic adjustment method for traction control of heavy-haul trains according to claim 1, characterized in that, When converting the decision instructions into corresponding physical actions, a dynamic relaxation transition strategy and a safety limiting mechanism must be satisfied. The dynamic relaxation transition strategy is as follows: when the gradient changes abruptly, the main locomotive of the heavy-haul train outputs traction force in advance, and the slave locomotive follows according to an exponential curve. The safety limiting mechanism is as follows: real-time monitoring of the actuator status; if the output deviates from the decision command by more than a preset standard, self-check and degraded operation are triggered.

17. A dynamic adjustment device for traction control of a heavy-haul train, used to implement the dynamic adjustment method for traction control of a heavy-haul train as described in any one of claims 1-16, characterized in that, include: The sensing module is used to collect real-time data on the operating status of heavy-haul trains, track environment, and coupler stress to obtain status data. The communication module is used to ensure real-time interaction between control commands and status data, and to build a redundant communication network; The decision module is used to generate decision instructions for dynamic allocation of traction force based on the state data and a multi-objective optimization strategy. The execution module is used to convert the decision command into a corresponding physical action based on the locomotive model of the heavy-haul train. The optimization module is used to collect real-time operating data of the heavy-haul train and upload it to the cloud digital twin, and to train deep reinforcement learning algorithm strategies offline in batches to generate new control rules.

18. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the dynamic adjustment method for traction control of heavy-haul trains as described in any one of claims 1 to 16.

Images