Heavy haul train automatic driving method, device and medium

Abstract

The application discloses a heavy-load train automatic driving method and device and a medium, relates to the field of train automatic driving, and comprises the following steps: establishing a reduced-order longitudinal dynamics model of a heavy-load train; constructing a coupler force identification model based on a segmented coupler coupling coefficient; obtaining actual line data of the heavy-load train, and substituting the actual line data into the reduced-order longitudinal dynamics model of the heavy-load train to obtain unmodeled dynamic terms corresponding to the actual line data; when a hybrid control model is used for automatic driving control of the heavy-load train, the unmodeled dynamic terms are pre-estimated and compensated to obtain an estimated value of the unmodeled dynamic terms; the hybrid control model is optimized based on the estimated value of the unmodeled dynamic terms, and the automatic driving control of the heavy-load train is performed based on the optimized hybrid control model, so that the modeling accuracy of the heavy-load train is improved, and the safe automatic driving of the heavy-load train is realized.

Description

Technical Field

[0001] This application relates to the field of automatic train operation, and in particular to a method, device and medium for automatic train operation of heavy-duty trains. Background Technology

[0002] In heavy-haul train control systems, time delays not only severely affect the control quality of the system but can also lead to safety accidents. In particular, the significant time delay effect of underactuated train control exacerbates the longitudinal dynamic coupling between vehicles, especially during speed regulation and braking, where high-amplitude longitudinal coupler forces are easily generated at the front and rear of the coupled locomotives.

[0003] Traditional multi-mass dynamics models for heavy-haul trains suffer from high computational complexity, making it difficult to meet the real-time computing requirements of onboard computers. This technical bottleneck leads to insufficient closed-loop control capabilities in train automatic driving systems. To address these issues and improve line transport capacity while ensuring operational safety, it is necessary to focus on exploring a data-driven-mechanism hybrid model that includes reduced-order longitudinal forces with unmodeled dynamics, and to formulate appropriate control strategies based on this model to verify its effectiveness.

[0004] In summary, there is an urgent need for a high-precision method, equipment, and medium for automatic driving of heavy-haul trains. Summary of the Invention

[0005] The purpose of this application is to provide a method, device and medium for automatic driving of heavy-haul trains, which can improve the modeling accuracy of heavy-haul trains and realize safe automatic driving of heavy-haul trains.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] Firstly, this application provides an automatic driving method for heavy-haul trains, including:

[0008] A reduced-order longitudinal dynamics model of a heavy-haul train is established based on the train parameters of the heavy-haul train.

[0009] Construct a coupler force identification model based on segmented coupler coupling coefficient;

[0010] Obtain the actual route data of heavy-haul trains and substitute the actual route data into the reduced-order longitudinal dynamics model of heavy-haul trains to obtain the unmodeled dynamic terms corresponding to the actual route data.

[0011] When using a hybrid control model for automatic control of heavy-haul trains, the unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms; the hybrid control model is optimized based on the estimated value of the unmodeled dynamic terms, and automatic control of heavy-haul trains is performed based on the optimized hybrid control model; the hybrid control model is constructed from the control force of the car, the coupler force identification model based on the segmented coupler coupling coefficient, the resistance experienced by the car, and the unmodeled dynamic terms.

[0012] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the above-described heavy-duty train automatic driving method.

[0013] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for automatic driving of heavy-duty trains.

[0014] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0015] This application provides a method, device, and medium for automatic driving of heavy-haul trains. It establishes a reduced-order longitudinal dynamics model of the heavy-haul train; constructs a coupler force identification model based on segmented coupler coupling coefficients; acquires actual track data of the heavy-haul train and substitutes this data into the reduced-order longitudinal dynamics model to obtain unmodeled dynamic terms corresponding to the actual track data; when using a hybrid control model for automatic driving control of the heavy-haul train, it performs prediction compensation for the unmodeled dynamic terms to obtain predicted values; it optimizes the hybrid control model based on the predicted values ​​of the unmodeled dynamic terms, and performs automatic driving control of the heavy-haul train based on the optimized hybrid control model. This application improves the modeling accuracy of the hybrid control model by predicting and compensating for the unmodeled dynamic terms and optimizing the hybrid control model based on the predicted values ​​of the unmodeled dynamic terms, effectively reducing the negative impact of insufficient model accuracy. This effectively reduces the longitudinal impact generated when the central locomotive enters the braking section, reduces the risk of decoupling of the central locomotive, and also reduces the wear rate of the coupler and extends the working life of the coupler. This effectively improves the transportation safety of heavy-haul trains and realizes safe automatic driving of heavy-haul trains. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the application environment of an automatic driving method for heavy-load trains according to an embodiment of this application.

[0018] Figure 2 This is a flowchart illustrating an embodiment of an automatic driving method for heavy-duty trains provided in this application.

[0019] Figure 3 A schematic diagram showing the comparison of convergence radii between the Northern Goshawk Optimization (NGO) algorithm and the Improved Northern Goshawk Optimization (INGO) algorithm provided in an embodiment of this application.

[0020] Figure 4 This is a schematic diagram of the Ingo-LSTM model calibration process provided in an embodiment of this application.

[0021] Figure 5 An estimation curve of the front-end delay compensation term in an INGO-LSTM model provided in an embodiment of this application; Figure 5 (a) in the figure is the estimated curve of the front-end delay compensation term from 0 to 2200 seconds. Figure 5 (b) is an enlarged view of the estimated curve of the front-end delay compensation term for 1380-1420 seconds. Figure 5 (c) is an enlarged view of the estimated curve of the front-end delay compensation term from 1715 to 1735 seconds.

[0022] Figure 6 This is a schematic diagram showing the comparison results of locomotive output under different models provided in an embodiment of this application; Figure 6 (a) in the figure shows the comparison results of the output of the central locomotive under different models. Figure 6 (b) shows the comparison results of the head locomotive output under different models.

[0023] Figure 7 This is a schematic diagram showing the stability comparison results of a control model provided in an embodiment of this application.

[0024] Figure 8 This is a schematic diagram showing the energy consumption comparison results of different model simulations provided in an embodiment of this application.

[0025] Figure 9 This is a schematic diagram of a hardware-in-the-loop simulation of heavy-load train operation provided in one embodiment of this application.

[0026] Figure 10 This is a schematic diagram of the coupler force variation curves of the front and rear couplers of the central locomotive under a fast terminal sliding mode control strategy, provided as an embodiment of this application.

[0027] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] The heavy-haul train automatic driving method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send the train parameters of the heavy-haul train to server 104. After receiving the train parameters, server 104 establishes a reduced-order longitudinal dynamic model of the heavy-haul train based on the train parameters, constructs a coupler force identification model based on the segmented coupler coupling coefficient, obtains the actual track data of the heavy-haul train, and substitutes the actual track data into the reduced-order longitudinal dynamic model of the heavy-haul train to obtain the unmodeled dynamic terms corresponding to the actual track data. When using a hybrid control model for automatic driving control of the heavy-haul train, the unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms. Server 104 can feed back the obtained estimated value of the unmodeled dynamic terms to terminal 102. In addition, in some embodiments, the heavy-haul train automatic driving method can also be implemented by the server 104 or the terminal 102 alone. For example, the terminal 102 can directly perform unmodeled dynamic term prediction compensation for the train parameters of the heavy-haul train, or the server 104 can obtain the train parameters of the heavy-haul train from the data storage system and perform unmodeled dynamic term prediction compensation for the train parameters of the heavy-haul train.

[0031] The terminal 102 can be, but is not limited to, various desktop computers and laptops. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0032] In one exemplary embodiment, such as Figure 2As shown, an automatic driving method for heavy-haul trains is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 204.

[0033] Step 201: Establish a reduced-order longitudinal dynamics model of the heavy-haul train based on the train parameters of the heavy-haul train.

[0034] Step 202: Construct a coupler force identification model based on the segmented coupler coupling coefficient.

[0035] Step 203: Obtain the actual line data of the heavy-haul train and substitute the actual line data into the reduced-order longitudinal dynamics model of the heavy-haul train to obtain the unmodeled dynamic terms corresponding to the actual line data.

[0036] Step 204: When using a hybrid control model for automatic control of heavy-haul trains, the unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms; the hybrid control model is optimized based on the estimated value of the unmodeled dynamic terms, and automatic control of heavy-haul trains is performed based on the optimized hybrid control model; the hybrid control model is constructed from the control force of the carriage, the coupler force identification model based on the segmented coupler coupling coefficient, the resistance experienced by the carriage, and the unmodeled dynamic terms.

[0037] By implementing steps 201 to 204 above, the hybrid control model is optimized based on the estimated values ​​of the unmodeled dynamic terms through prediction and compensation of the unmodeled dynamic terms. This improves the modeling accuracy of the hybrid control model and effectively reduces the negative impact of insufficient model accuracy. Consequently, it effectively reduces the longitudinal impact generated when the central locomotive enters the braking section, lowers the risk of decoupling of the central locomotive, and also reduces the wear rate of the coupler, extending its service life. This effectively improves the transportation safety of heavy-haul trains and enables safe automatic driving of heavy-haul trains. The proposed hybrid control model performs better in terms of accuracy, control force output, and control stability, thereby reducing longitudinal impulses of the coupler and reducing energy consumption to a certain extent. This application provides a theoretical basis for accurate modeling and safe control of heavy-haul trains.

[0038] The train parameters of the heavy-haul train include the car mass coefficient, car acceleration, car speed, car displacement, car aerodynamic braking force, car resistance, coupler system elastic coupling coefficient, coupler system damping coupling coefficient, time coefficient, and model delay. Based on the train parameters of the heavy-haul train and the dynamic underactuation that occurs during the operation of the heavy-haul train, a reduced-order longitudinal dynamic model of the heavy-haul train is established by combining dynamic simulation and underactuation level analysis.

[0039] Specifically, this embodiment first introduces a data-driven reduced-order identification method for the control model of long and heavy haul trains, which reduces the order of the traditional longitudinal dynamics model of heavy haul trains. The longitudinal dynamics model of heavy haul trains in this embodiment can truly map the running states of the 1st, 16th, 32nd, 48th, 55th, 64th, 80th, and 96th carriages of the heavy haul train body, so as to reflect the running conditions of the entire train formation. While maintaining the coupling characteristics between the locomotive and the carriages, this simplified model effectively reduces the system size, and the calculation efficiency is increased by more than 10 times. The reasons for selecting the 1st, 16th, 32nd, 48th, 55th, 64th, 80th, and 96th carriages are shown in the grouped parameter table of the underactuated model of heavy haul trains in Table 1. The selected carriages for the divided nine-particle model are highly representative of direct drive, semi-underactuated, and underactuated.

[0040] In this embodiment, by combining dynamic simulation and underactuated level analysis, the train formation is divided into three hierarchical groups (UAD, Underactuated Degree, which refers to the degree of underactuation. The lower the value, the lower the degree of underactuation and the stronger the control ability): UAD is defined by delay. When the delay is less than 50 ms, the degree of underactuation is within the range of 0 < UAD < 0.3; when the delay is between 50 - 150 ms, the degree of underactuation is within the range of 0.3 ≤ UAD < 0.7; when the delay exceeds 150 ms, the underactuated range is UAD ≥ 0.7.

[0041] (1) Direct drive layer: It has full traction / braking capabilities, a low degree of underactuation (UAD < 0.3), and is composed of a single locomotive or a locomotive multi-car cluster. The power unit can independently adjust the traction / braking force, covering the core degrees of freedom of the train. The response time is short, and the control command is almost executed in real time (the delay is negligible or extremely low), ensuring dynamic stability. Through multi-locomotive cooperation, the redundant control ability is enhanced, reducing the risk of single-point failures. It has the characteristics of high control freedom, low delay, and high reliability.

[0042] (2) Semi-underactuated layer: It shows medium underactuation (0.3 ≤ UAD < 0.7), with a response delay of about 150 milliseconds, and is completely composed of trailers near the direct control layer. It cannot actively output traction force, but can adjust the braking force through the braking system and needs to rely on upper-layer commands for coordination. The 150-millisecond response delay may cause fluctuations in the coupler force or lag in dynamic response. It can only adjust some degrees of freedom (such as braking force distribution, but cannot independently adjust the traction force).

[0043] (3) Underactuated layer: Requires high underactuation (UAD≥0.7) and indirect coordinated control with a delay exceeding 150 milliseconds, consisting only of trailers connected to the semi-underactuated layer. Traction / braking forces cannot be applied directly; adjustment is achieved solely through coupler force transmission and indirect commands from the upper layer. The delay exceeding 150 milliseconds leads to a mismatch between control commands and real-time conditions, easily causing sudden changes in coupler force or the risk of coupler breakage. Control inputs are far fewer than the system's degrees of freedom (e.g., adjusting the dynamics of the rear trailer solely through the front locomotive).

[0044] For heavy-haul trains using a "1+54+1+54" configuration (consisting of two tandem locomotive units of 1 locomotive + 54 carriages), the mechanical load of the intermediate locomotive is retained. The stability of this locomotive determines the collective safety control performance of the entire trainset. A simplified grouping scheme for the mechanical model is detailed in Table 1, achieving computational tractability while maintaining basic dynamic coupling.

[0045] Table 1 Grouping Parameter Table for Underactuated Model of Heavy-Duty Train

[0046]

[0047] Based on this, a longitudinal dynamic model of a heavy-haul train with nine mass points is established, which realistically maps the operating state of the 1st, 16th, 32nd, 48th, 55th, 64th, 80th, and 96th cars of the heavy-haul train, thus reflecting the operating status of the entire train formation. The reduced-order longitudinal dynamic model of the heavy-haul train is shown in expression (1):

[0048] (1);

[0049] in, Indicated as the heavy-haul train number The mass coefficient of each carriage =1, 2, 3, ..., 9; Represented as the first The acceleration of each carriage; Represented as the first The speed of each carriage; Represented as the first The displacement of each carriage; t For time coefficient; Delay the model; Represented as the first The locomotive output force or air braking force of each carriage; is the elastic coupling coefficient of the coupler system; c is the damping coupling coefficient of the coupler system. Represented as the first The resistance experienced by each carriage.

[0050] Furthermore, expression (1) in this embodiment can be simplified to matrix form, i.e., expression (2):

[0051] (2);

[0052] in, This is the mass coefficient matrix of the carriage. Indicates the diagonal symbol; For the air braking force matrix of the carriage; This represents the resistance matrix experienced by the carriage. Here is the displacement matrix of the carriage; The speed matrix of the carriage; The acceleration matrix of the carriage; This represents the elastic coupling coefficient matrix of the coupler system. This is the damping coupling coefficient matrix of the coupler system.

[0053] A coupler force identification model based on segmented coupler coupling coefficients is constructed, specifically including: defining the relative displacement and relative velocity between the cars as trigger variables for coefficient changes, determining the segmented coupler coupler coefficients; and constructing a coupler force identification model based on the segmented coupler coupler coefficients.

[0054] The relative displacement and relative velocity between carriages are defined as trigger variables for coefficient changes, and a coupler force identification model based on the segmented coupler coupling coefficient is constructed. In this embodiment, the coupler force identification model based on the segmented coupler coupling coefficient is constructed according to the complex nonlinear characteristics of coupler force during train start-up, speed adjustment, or braking.

[0055] It should be noted that the coupler is a key factor in connecting vehicle units and transmitting mechanical motion, and its constraint relationship is crucial for the analysis of system dynamics. Considering the complex nonlinear characteristics of coupler forces during train startup, speed regulation, or braking, a coupler force identification model based on segmented coupler coupling coefficients is invented. Analysis based on a large amount of measured data shows that when the deformation of the coupler buffer device reaches a certain level, the spring damping coefficient will also change accordingly.

[0056] In this embodiment, the relative displacement and relative velocity between the carriages are defined as trigger variables for coefficient changes, thereby defining the coupler coefficients, as follows: Elastic phase: when At the elastic limit, the spring constant is equal to the linear elastic stiffness coefficient of the spring. Plastic stage: when When the spring constant is the spring's plastic deformation stiffness coefficient, This reflects the nonlinear stiffness jump; the nonlinear damping coefficient cc ( ) is related to relative velocity, expressed as: cc ( )= + · ,in, It is the initial damping. It is a velocity-sensitive damping coefficient, which enables the damping to change dynamically with the absolute value of the relative velocity.

[0057] The nonlinear coupler force mathematical model, namely the coupler force identification model based on the segmented coupler coupling coefficient, is expressed as follows:

[0058] (3);

[0059] in, For the first Section and the +1 car coupler force; Indicates the first Section and the +1 relative displacement between carriages; Indicates the first Section and the +1 relative speed between carriages; This represents the linear stiffness coefficient of the spring; Indicates the stiffness coefficient of a spring under plastic deformation; Indicates initial damping. This represents the velocity-sensitive damping coefficient; sgn() is the sign bit function; Indicates the elastic limit; Indicates the preload within the operating range; Indicates the maximum tensile force of the coupling; This indicates the maximum compressive force of the coupling.

[0060] Actual track data of heavy-haul trains is measured and acquired. This data is then substituted into a reduced-order longitudinal dynamics model of the heavy-haul train to obtain the unmodeled dynamic terms at different times. Specifically, the measured actual track data is substituted into the reduced-order longitudinal dynamics model of the heavy-haul train, and differential operations are performed to obtain the unmodeled dynamic terms of the hybrid control model at different times. The unmodeled dynamic terms are used to determine the inputs to the control system, the train's operating status, and the corresponding unmodeled dynamic values ​​at different times. Actual track data for heavy-haul trains includes speed, displacement, resistance, and control force during actual train operation.

[0061] It should be noted that in large-scale train control systems, the existence of delay will seriously affect the control quality of the system. This embodiment compensates for the delay effect in the traditional longitudinal dynamics model. The effect of train control lag varies under different operating conditions, and it is difficult to describe the compensation amount of the control model at different times through mechanism. However, the real data of actual train operation (actual line data), including speed information, displacement information, control force, resistance and other information, can be obtained through measurement. Then, the measured line data is substituted into expression (2) and the difference operation is performed on both sides of the equation to obtain the unmodeled dynamic terms of the hybrid control model at different times. Thus, the time ( t The inputs to the train control system, train operation status data, and corresponding unmodeled dynamic values.

[0062] The hybrid control model for the heavy-haul train design in this embodiment is as follows:

[0063] (4);

[0064] in, Indicates the hybrid control model in time t Unmodeled dynamic values.

[0065] Based on step 203, the dataset for training the Long Short-Term Memory (LSTM) neural network model can be obtained.

[0066] Before inputting vehicle operation data into the trained LSTM model for prediction and compensation to obtain the predicted value of the unmodeled dynamic terms, the heavy-haul train automatic driving method also includes the LSTM model training process, with the specific steps as follows:

[0067] Construct a dataset; the dataset includes actual line data and unmodeled dynamic terms corresponding to the actual line data output by the reduced-order longitudinal dynamics model of heavy-load trains;

[0068] Using actual line data as input and the unmodeled dynamic terms corresponding to the actual line data output by the reduced-order longitudinal dynamics model of heavy-haul trains as output, the parameters of the LSTM model are optimized using the improved Northern Eagle algorithm to obtain a trained LSTM model.

[0069] To obtain heavy-haul train operation data, the InGO-LSTM prediction and correction algorithm is used to predict and compensate for unmodeled dynamic terms. The process is as follows: Train the LSTM model to obtain the optimal LSTM model parameters; establish an unmodeled dynamic term compensation model using the optimal LSTM model parameters, which is the trained LSTM model; output the compensation result using the unmodeled dynamic term compensation model, which is the predicted value of the unmodeled dynamic terms. The compensation result is used to compensate for the delay in the heavy-haul train control system.

[0070] The training of the LSTM model includes: using the actual operating speed, displacement, resistance, and control force of the heavy-haul train as training inputs, and using the unmodeled dynamic terms calculated by the reduced-order longitudinal dynamics model of the heavy-haul train as training outputs; initializing the INGO parameters and LSTM model parameters, and simultaneously initializing the position of the Northern Eagle within the range of LSTM model parameter values; and optimizing the LSTM model parameters based on the INGO algorithm to obtain the optimal LSTM model parameters.

[0071] INGO parameters include population size and maximum number of iterations, while LSTM model parameters include the number of hidden layer nodes, initial learning rate, and regularization coefficient.

[0072] An LSTM model optimized using an improved NGO algorithm is employed to predict and compensate for unmodeled dynamic terms in the hybrid control model, resulting in a high-precision hybrid control model. By optimizing the hybrid control model, the model accuracy is improved, enabling automatic driving of heavy-haul trains.

[0073] To reduce the impact of control model errors on the control process, this embodiment designs a prediction compensation model based on Ingo-LSTM. By obtaining actual track data, including speed, displacement, resistance, and control force, it predicts the unmodeled dynamic terms in the hybrid control model. The design of the unmodeled dynamic term compensation model is described below using the unmodeled dynamic terms of the leading locomotive as an example:

[0074] (5);

[0075] in, express k The estimated value of unmodeled dynamic terms in the timetable train operation control system. express k The mapping relationship between actual time-of-flight route data and estimated values ​​of unmodeled dynamic items.

[0076] Training an LSTM model requires establishing the intrinsic relationship between vehicle operation data and the unmodeled dynamics in the hybrid control model. This application utilizes real-world operation data of heavy-haul trains in a specific location, inputting the collected actual track data into the longitudinal dynamics model of the heavy-haul train to obtain accurate values ​​of the unmodeled dynamics at each time step. Then, using the actual track data as input and the corresponding unmodeled dynamics in the reduced-order longitudinal dynamics model of the heavy-haul train as the model output, a dataset is constructed to train the model. The trained LSTM model can accurately map the intrinsic relationship between the operation data and the unmodeled dynamics of the model, thereby enabling online prediction of the unmodeled dynamics of the control model during heavy-haul operation.

[0077] In this embodiment, the LSTM model includes an INGO-LSTM neural network. The INGO-LSTM neural network utilizes an improved NGO algorithm to optimize three parameters of the LSTM model: the number of hidden layer nodes, the initial learning rate, and the regularization coefficient, in order to accelerate the convergence speed of the neural network. The forward execution process of the LSTM model is as follows:

[0078] (6);

[0079] in, Here, sigmoid is the function, and tanh is the hyperbolic cosine function; Forgotten Gate; These are the first weight matrix, the second weight matrix, and the bias vector of the forget gate, respectively; For input gates; These are the first weight matrix, the second weight matrix, and the bias vector of the input gate, respectively; For output gate; These are the first weight matrix, the second weight matrix, and the bias vector of the output gate, respectively. for The state variables of the memory unit at any given time; for The state variables of the memory unit at any given time; For vector scalar product, for Memory cells of a moment; for The output of the forget gate; for The output of the input gate is always being input; These are the first weight matrix, the second weight matrix, and the bias vector of the memory cell, respectively. for Input information at any time; for Output information at any given moment.

[0080] It should be noted that the NGO algorithm mainly consists of two stages: the first stage is to determine the location of the eagle and the target; the second stage is for the eagle to quickly move towards the target to launch an attack, chase and hunt the prey.

[0081] In this embodiment, the NGO algorithm has good solution accuracy and stability, but it still has the following limitations: ① In the algorithm initialization stage, the population is randomly initialized, the initial solution is unevenly distributed, and the population diversity is insufficient; ② In the second stage, the convergence radius R decreases linearly, the attack range is small, and it takes a long time to capture prey in the early stage of iteration, which cannot handle complex nonlinear problems in the actual optimization process.

[0082] To address the above shortcomings, this embodiment introduces the Tent chaotic mapping in the initialization phase of the improved NGO algorithm. The Tent chaotic mapping is a chaotic mapping method with good uncertainty and randomness. Compared to other chaotic mapping methods, the Tent mapping has a more uniform distribution function in its mapping space. Its expression is:

[0083] (7);

[0084] in, For the first Only the location of the northern eagle; For the first Only the location of the northern eagle; This is a user-defined parameter used to select the mapping range.

[0085] The formula for the convergence radius of the Eagle attack was redesigned in the second phase as follows:

[0086] (8);

[0087] in, The convergence radius of the eagle's attack. It is a sine function; Pi; This represents the current iteration number; This represents the maximum number of iterations.

[0088] Specifically, the attack convergence radius of the NGO algorithm and the improved NGO algorithm is as follows: Figure 3 As shown, compared to the initial NGO's attack radius, the new convergence radius expands the individual's attack range. Furthermore, it exhibits a non-linear trend of slow initial decrease followed by a sharp decrease later as the number of iterations increases. This indicates that the individual exerts a significant effort in exploring the entire search space in the early stages, improving the algorithm's global optimization capability; the later stages imply faster convergence. Compared to the original NGO algorithm, the improved NGO algorithm achieves a better balance between global and local search capabilities.

[0089] Simultaneously, after sampling the new convergence radius, to prevent the algorithm from getting too fast in the later stages and getting trapped in local optima, the spiral predation behavior of WOA is combined with the hunting model of the second stage of the NGO algorithm. This allows the eagle to gradually approach the prey in a spiral manner, improving its local search capability. The improved hunting model of the second stage of the NGO algorithm is changed to:

[0090] (9);

[0091] in, For the second phase Only the Northern Eagle A new position in the dimension; It is a natural exponential function; Logarithmic spiral shape constant coefficient; It is a random number in the range [-1, 1]. It is a cosine function; For the first Only the northern eagle in the first The position of the dimension; It is a random number in [-1, 1].

[0092] INGO-LSTM model such as Figure 4 As shown, the specific steps are as follows:

[0093] Step 1: The actual route data of the heavy-haul train (speed, displacement, resistance, and control force) is used as the training input, and the unmodeled dynamic terms calculated by the mechanistic model are used as the training output. Figure 4 The mechanism model in this context refers to the reduced-order longitudinal dynamics model of a heavy-haul train.

[0094] Step 2: Initialize the INGO algorithm search dimension to 3, with the search targets being the number of hidden layer nodes in the LSTM, the initial learning rate, and the regularization coefficient. Set the maximum number of iterations to 500. Simultaneously, initialize the position of the Northern Eagle within the three LSTM model parameters to be optimized. The number of hidden layer nodes is typically between 32 and 512. The initial learning rate is generally taken as 1e-5 (0.00001) to 1e-1 (0.1), with commonly used values ​​being 1e-4 to 1e-2; the regularization coefficient is typically between 1e-5 (0.00001) and 1e-2 (0.01).

[0095] Step 3: Based on the optimization steps of the INGO algorithm, and combined with the training samples of heavy-haul train data, train the LSTM model and calculate the individual (location of the Northern Goshawk) in the current population, that is, the number of hidden layer units, learning rate and normalization coefficients in the LSTM.

[0096] Step 4: Determine the optimal solution for the current population based on the fitness value of each individual, and save the optimal solution and fitness value. The fitness function, Fitness = -MSE, is defined using the mean squared error (MSE) as the evaluation metric, and the individual with the highest fitness value is the optimal solution.

[0097] (10);

[0098] in, It is the actual value. These are the predictions from the LSTM model. This represents the number of training samples.

[0099] Step 5: Determine if the iteration termination condition has been met. If yes, display the optimal solution, output the optimal model parameters, and complete the LSTM hyperparameter optimization selection. The optimal LSTM model parameters are: 49 hidden layer nodes; initial learning rate of 0.006; and regularization coefficient of 0.0012. If no, return to Step 3 until the iteration termination condition is met. The iteration termination condition can be that the current iteration count has reached the maximum iteration count.

[0100] Step 6: Establish a compensation model for unmodeled dynamic terms using the optimal model parameters, and use the output results of the trained LTSM model to compensate for the delays present in the mechanism model of heavy-load trains.

[0101] In step 204 above, the unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms. Specifically, this includes: inputting the actual line data into the trained LSTM model for estimation and compensation to obtain the estimated value of the unmodeled dynamic terms.

[0102] By using the above methods to predict and compensate for uncertainties in the control system model, a high-precision system control model can be obtained, thereby eliminating the negative impact of modeling errors on control performance and, to some extent, compensating for the limitations of the control algorithm. The train control model based on Ingo-LSTM prediction, i.e., the reduced-order hybrid control model, is as follows:

[0103] (11);

[0104] in, This represents the estimated value of the unmodeled dynamic terms in the hybrid control model of heavy-haul trains.

[0105] The data simulation is performed using the method provided in this embodiment, as follows: In this embodiment, the train adopts a "1+54+1+54" formation, with the electric locomotive model being HXD1 and the freight car model being C80. The parameters of the entire freight train are shown in Table 2.

[0106] Table 2 Parameters of the entire freight train

[0107]

[0108] To determine the effectiveness of the estimated compensation model in this embodiment, the instability index of the estimated compensation model should be determined based on the functional nature of the system and the importance of the performance requirements of concern. This application uses unmodeled compensation terms. As a performance indicator, To predict the set of states for the stable R of the compensation model, the corresponding failure state coefficient criterion is as follows:

[0109] (12);

[0110] in, The estimated value is for the compensation term that was not modeled at that time; the threshold range for the performance metric is: =100N.

[0111] To verify the effectiveness of the estimated compensation model proposed in this embodiment, a speed tracking curve was set based on the actual operating speed of a train in a certain location, including acceleration, coasting, and braking conditions of the heavy-haul train, with a cumulative simulation time of 2260 seconds. It should be noted that: based on the collected speed and control force data of the heavy-haul train during operation, the time delay during the test run of the train in the laboratory hardware-in-the-loop simulation platform was comprehensively analyzed to obtain the appropriate lag time. At 0.4s, the value best matches the field characteristics of the heavy-haul train control process and can relatively effectively describe the time delay of the train in actual operation. By comparing whether the difference between the prediction results of the unmodeled dynamic terms by INGO-LSTM and the ideal values ​​in the model exceeds a given range, the stability of the system can be determined, and the proportion of instability states in the operation of the entire large train control system can be recorded. Table 3 shows the instability indices of hybrid models combining LSTM, NGO-LSTM, and INGO-LSTM network models, respectively. The closer the failure index is to 0, the more stable the prediction compensation model is.

[0112] Table 3 Comparison of Model Stability Indicators

[0113]

[0114] It is evident that the LSTM model optimized using the INGO algorithm proposed in this application exhibits good stability.

[0115] Figure 5 The estimated curve of the locomotive delay compensation term in the INGO-LSTM model is shown. Clearly, the INGO-LSTM effectively predicts the locomotive's time delay component, thus reconstructing a hybrid control model for the control system of long-formation heavy-haul trains.

[0116] To verify the significance of the results of this embodiment for the stable operation of heavy-haul trains, a stability evaluation function for the operation of the central locomotive was obtained using a weighted method. The formula for the stability evaluation function is as follows:

[0117] (13);

[0118] in, For stationarity evaluation function; This refers to the operating speed of the locomotives in the central region. To accelerate the operation of locomotives in the central region; Traction / braking impulse; weighting factor .

[0119] Train traction energy consumption, as a key criterion for evaluating control performance, should be considered when verifying the results proposed in this application. The energy consumption of heavy-haul trains during operation mainly consists of two parts: first, the energy consumed by locomotive traction; second, the energy consumed by train auxiliary equipment (including lighting, onboard electrical equipment, etc.). Since the energy consumption of auxiliary equipment on heavy-haul trains is relatively fixed, it is not the focus of this discussion. Energy consumed during locomotive control... It can be represented as:

[0120] (14);

[0121] in, For the traction force of high-speed trains; The speed at the current moment; For train motor efficiency; The conversion efficiency when the train uses regenerative braking; The braking force generated when a high-speed train is about to arrive at the station; Energy consumption for train auxiliary power per unit time; These are the start and end times of the controlled time period, respectively.

[0122] When control time delay exists, under the fast terminal sliding mode control strategy, simulations of heavy-haul trains are conducted using both the traditional model and the model proposed in this application. The comparison results of locomotive output under different models are as follows: Figure 6 As shown, when using the traditional model, the control forces of both the leading locomotive and the middle locomotive exhibit significant oscillations. However, when using the model presented in this paper, the control time delay is effectively suppressed, and the locomotive control forces become relatively stable.

[0123] Under the same fast terminal sliding mode control strategy, the stability of heavy-haul train simulation operation was compared using a multi-mass longitudinal dynamics model and the model proposed in this application. The stability comparison results of the control models are as follows: Figure 7 As shown, Figure 7The model proposed in this invention refers to a hybrid control model. Based on the stability evaluation function determined by equation (13), a specific analysis of stability shows that the control strategy using this application's model has an average evaluation function value of 0.018, indicating relative stability. In contrast, the control strategy without delay compensation has an average evaluation function value of 0.034 and exhibits significant impulses during control. It can be concluded that this application's model can improve controller performance by approximately 47%.

[0124] The energy consumption curves of the leading locomotive during the entire vehicle operation period, under the same fast terminal sliding mode control strategy as the model in this application and the multi-mass longitudinal dynamics model, are compared. The results are as follows: Figure 8 As shown, during the initial traction phase of the journey, neither compensating for the unmodeled dynamic terms in the control model nor affecting the train's traction energy consumption remains a concern. However, during the constant speed and braking phases, insufficient modeling accuracy of the multi-mass longitudinal dynamics model leads to significant oscillations in the locomotive's output force, thereby increasing energy consumption during train operation. At the same preset tracking speed, the total energy consumption of the locomotive simulation using the multi-mass longitudinal dynamics model is 2.26 × 10⁻⁶. 5 (kW), while the total energy consumption of the model locomotive simulation in this application is 2.15×10 5 (kW), indicating that the hybrid control model of this application improves the control performance of the controller and can reduce the locomotive energy consumption by 4.87%.

[0125] Hardware-in-the-loop simulation diagram of heavy-haul train operation is shown below. Figure 9 As shown, it includes the overall coupler force distribution of a heavy-haul train, which can more intuitively describe the control effect of the model designed in this application, and demonstrate the simulated operation effect of the train between Shenchi South Station and Longgong Station on the Shuohuang Railway. In the enlarged detail, from top to bottom, the first curve is coupler position 64, the second curve is coupler position 80, the third curve is coupler position 1, the fourth curve is coupler position 96, the fifth curve is coupler position 16, the sixth curve is coupler position 32, and the seventh curve is coupler position 48. It can be seen that the change of the key coupler coupling force is relatively stable and the order is clear, which is conducive to reducing wear and extending the service life of the coupler.

[0126] For the traction control system of a 1+1 heavy-haul locomotive, when the train is under normal traction conditions, the front coupler of the middle locomotive is mainly in a compressed state, while the rear coupler is in a stretched state. When the train enters the braking section, the middle locomotive decelerates first relative to the front and rear carriages, resulting in a rapid change in coupler force and generating a large longitudinal impact. The purpose of this application is not only to ensure the overall stability of the heavy-haul train during operation, but also to minimize the longitudinal impact of the middle locomotive under braking conditions, i.e., to reduce the maximum coupler force during braking, thereby reducing the risk of decoupling of the middle locomotive.

[0127] To verify the significance of this application's results in reducing the longitudinal impact of the central locomotive, the coupler data of the central locomotive under manual control on the freight line between Shenchi South Station and Longgong Station were compared with the coupler data of the central locomotive under the fast terminal sliding mode control strategy of the model in this application. Table 4 shows the coupler information of the central locomotive in the four braking sections of this section under manual control.

[0128] Table 4. Locomotive Braking Information for the Middle Section Between Shenchi South Station and Longgong Station

[0129]

[0130] Figure 10 The curves showing the coupler force variation of the front and rear couplers of the central locomotive under the fast terminal sliding mode control strategy of this application model are shown. Figure 10 In the braking sections 1 to 4, the maximum hooking forces of the central locomotive coupler were 494.2 kN, 902.2 kN, 577.1 kN, and 687.4 kN, respectively. Compared with manual operation, the maximum hooking forces decreased by 10.7%, 26.2%, 9.5%, and 8.2%. The maximum hooking forces of the central locomotive coupler were 239.8 kN, 447.4 kN, 279.5 kN, and 652.9 kN, respectively. Compared with manual operation, the maximum hooking forces decreased by 13.2%, 3.2%, 5.0%, and 7.1%.

[0131] This is sufficient to prove that the model in this application can effectively reduce the negative impact caused by insufficient model accuracy, thereby effectively reducing the longitudinal impact generated when the locomotive in the middle enters the braking section, reducing the risk of the locomotive in the middle decoupling, and also reducing the wear rate of the coupler and extending the working life of the coupler, which effectively improves the safety of heavy-haul train transportation.

[0132] This application also provides an application scenario in which the above-mentioned heavy-haul train automatic driving method is applied. Specifically, the heavy-haul train automatic driving method provided in this embodiment can be applied in a heavy-haul train automatic driving scenario. The heavy-haul train automatic driving scenario includes a parameter acquisition stage, an unmodeled dynamic term prediction and compensation link, and an automatic driving control stage. The train parameters of the heavy-haul train enter the unmodeled dynamic term prediction and compensation link from the parameter acquisition stage. Through human-machine collaboration, the predicted value of the corresponding unmodeled dynamic term is obtained and then enters the downstream automatic driving control stage. The heavy-haul train automatic driving method provided in this embodiment belongs to the unmodeled dynamic term prediction and compensation link. Specifically, in the process of unmodeled dynamic term prediction and compensation link for heavy-haul trains, a reduced-order longitudinal dynamic model of the heavy-haul train can be established based on the train parameters of the heavy-haul train. A coupler force identification model based on the segmented coupler coupling coefficient can be constructed to obtain the actual line data of the heavy-haul train. The actual line data is then substituted into the reduced-order longitudinal dynamic model of the heavy-haul train to obtain the unmodeled dynamic term corresponding to the actual line data. When using a hybrid control model for heavy-haul train automatic driving control, the unmodeled dynamic term is predicted and compensated to obtain the predicted value of the unmodeled dynamic term.

[0133] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 11 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores automatic train operation data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an automatic train operation method.

[0134] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0135] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0136] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0138] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0139] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0140] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for automatic driving of heavy-haul trains, characterized in that, The heavy-haul train automatic driving method includes: A reduced-order longitudinal dynamics model of a heavy-haul train is established based on the train parameters of the heavy-haul train. A coupler force identification model based on segmented coupler coupling coefficients is constructed; the coupler force identification model based on segmented coupler coupling coefficients is represented as follows: ； in, For the first Section and the +1 car coupler force; Indicates the first Section and the +1 relative displacement between carriages; Indicates the first Section and the +1 relative speed between carriages; This represents the linear stiffness coefficient of the spring; Indicates the stiffness coefficient of a spring under plastic deformation; Indicates initial damping. This represents the velocity-sensitive damping coefficient; sgn() is the sign bit function; Indicates the elastic limit; Indicates the preload within the operating range; Indicates the maximum tensile force of the coupling; Indicates the maximum compressive force of the coupling; Obtain the actual route data of heavy-haul trains and substitute the actual route data into the reduced-order longitudinal dynamics model of heavy-haul trains to obtain the unmodeled dynamic terms corresponding to the actual route data. When using a hybrid control model for automatic control of heavy-haul trains, the unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms; the hybrid control model is optimized based on the estimated value of the unmodeled dynamic terms, and automatic control of heavy-haul trains is performed based on the optimized hybrid control model; the hybrid control model is constructed from the control force of the car, the coupler force identification model based on the segmented coupler coupling coefficient, the resistance experienced by the car, and the unmodeled dynamic terms.

2. The automatic driving method for heavy-haul trains according to claim 1, characterized in that, The train parameters of the heavy-haul train include the car mass coefficient, car acceleration, car speed, car displacement, car aerodynamic braking force, car resistance, coupler system elastic coupling coefficient, coupler system damping coupling coefficient, time coefficient, and model delay.

3. The automatic driving method for heavy-haul trains according to claim 1, characterized in that, A coupler force identification model based on segmented coupler coupling coefficients is constructed, specifically including: The relative displacement and relative velocity between the carriages are defined as the trigger variables for coefficient changes, and the coefficients of the segment coupler are determined. A coupler force identification model based on the segmented coupler coupling coefficient is constructed.

4. The automatic driving method for heavy-haul trains according to claim 1, characterized in that, The unmodeled dynamic terms are estimated and compensated to obtain the estimated value of the unmodeled dynamic terms, specifically including: The actual line data is input into the trained LSTM model for prediction and compensation, and the predicted value of the unmodeled dynamic term is obtained.

5. The heavy-haul train automatic driving method according to claim 4, characterized in that, Before inputting vehicle operation data into a trained LSTM model for prediction and compensation to obtain a predicted value for the unmodeled dynamic term, the heavy-haul train automatic driving method further includes: Construct a dataset; the dataset includes actual line data and unmodeled dynamic terms corresponding to the actual line data output by the reduced-order longitudinal dynamics model of heavy-load trains; Using actual line data as input and the unmodeled dynamic terms corresponding to the actual line data output by the reduced-order longitudinal dynamics model of heavy-haul trains as output, the parameters of the LSTM model are optimized using the improved Northern Eagle algorithm to obtain a trained LSTM model.

6. The automatic driving method for heavy-haul trains according to claim 5, characterized in that, The parameters of an LSTM model include the number of hidden layer nodes, the initial learning rate, and the regularization coefficient.

7. The automatic driving method for heavy-haul trains according to claim 1, characterized in that, The actual track data includes the speed, displacement, resistance, and control force of the actual train during operation.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the heavy-haul train automatic driving method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the heavy-load train automatic driving method according to any one of claims 1-7.

Images