A multi-stage cooperative control method and system for heavy-load combined trains

By acquiring train status and track data, calculating total demand force based on gradient prediction, and constructing a model predictive control framework, the traction and braking force commands of heavy-haul combined trains are optimized. This solves the problem of balancing speed safety and robustness when heavy-haul combined trains operate on complex slopes, achieving higher operational safety and stability.

CN123104745APending Publication Date: 2026-11-17ZHUZHOU CSR TIMES ELECTRIC CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611619466.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When heavy-haul combined trains run on long downhill slopes and complex undulating slopes, it is difficult to balance speed safety, longitudinal stability and execution robustness. Existing technologies do not fully consider the differences in traction or braking force between the master locomotive and the slave locomotive at different gradients, resulting in excessive or insufficient force. Furthermore, air braking has not been optimized for nonlinear, time-delay and hysteresis characteristics.

Method used

By acquiring train status and track data, the total demand force is calculated based on train force balance and gradient prediction, and differential allocation is performed to construct a model predictive control framework. This optimizes traction, electric braking force, and air braking force commands, suppresses longitudinal impulses, and improves robustness.

Benefits of technology

It achieves a balance between speed safety, longitudinal stability, and robust operation when running on complex slopes, significantly improving the operational safety and reliability of heavy-haul combined trains.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104745A_ABST
    Figure CN123104745A_ABST
Patent Text Reader

Abstract

The application provides a kind of multi-stage cooperative control method and system of heavy load combined train, it is related to heavy load train control field.The method comprises: obtaining train state, line data and master-slave control locomotive state and respective real-time slope;Vehicle-ground speed safety layer is based on stress balance and slope difference calculation and distribution total demand force, output safety boundary force and adjustment margin;Vehicle-car smooth optimization layer minimizes the objective of car coupler force at slave locomotive, constructs model predictive control under the constraint of communication delay and power asynchronization, optimizes traction / electric braking instruction;Air-electricity robust execution layer determines execution error according to air brake inverse model and actual brake cylinder pressure, corrects electric force and feedback.The application can cooperatively improve speed safety, longitudinal smoothness and execution robustness under complex slope.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of heavy-haul train control, and in particular to a multi-level cooperative control method and system for heavy-haul combined trains. Background Technology

[0002] Heavy-haul combined trains are long, heavy, and have high inertia. When running on long downhill slopes and complex undulating inclines, their dynamic behavior is complex, making it difficult to simultaneously ensure speed safety, longitudinal stability, and execution robustness. Existing technologies often simplify the track gradient to an average gradient for traction calculations, failing to fully consider the significant differences in traction or braking forces required when the master and slave locomotives are simultaneously at different gradients. This can easily lead to excessive or insufficient force in some locomotives, and even cause motor overload, wheel-rail adhesion imbalance, and speed runaway. In vehicle-to-vehicle cooperative control, communication delay is usually only compensated for as a simple signal lag, and is not incorporated into the optimization framework aimed at minimizing coupler force. Furthermore, it is assumed that the air brakes of the master and slave locomotives are consistent, without utilizing independent and differentiated air braking force distribution to suppress and mitigate longitudinal surges caused by asynchrony. Air-electric coordination is mostly limited to logic switching, failing to consider air braking as an execution link with nonlinear, time-delayed, and hysteresis characteristics, and lacking a local real-time compensation mechanism.

[0003] Therefore, how to ensure speed safety, longitudinal stability, and execution robustness simultaneously during the automatic driving process of heavy-haul combined trains, and achieve multi-objective collaborative optimization, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide a multi-level cooperative control method and system for heavy-haul combined trains, which can solve the problems of difficulty in coordinating and optimizing speed safety, longitudinal stability and execution robustness in the existing multi-level cooperative control of heavy-haul combined trains on complex slopes.

[0005] To address the aforementioned technical problems, this application provides a multi-level cooperative control method for heavy-haul combined trains, comprising: The system acquires train status data, track data, and real-time gradients of the master locomotive and slave locomotive. The vehicle-to-ground speed safety layer calculates the total train demand force based on the train force balance and forward gradient prediction, using the status data, track data, and real-time gradients of the master and slave locomotives. It then performs differential allocation of the total train demand force based on the gradient difference between the master and slave locomotives, outputting the allocation result. The allocation result includes the master locomotive safety boundary force, the slave locomotive safety boundary force, the master locomotive adjustment margin, and the slave locomotive adjustment margin. The vehicle-to-vehicle stability optimization layer receives the allocation result and, with the goal of minimizing the coupler force at the slave locomotive and constrained by communication delay and power asynchrony characteristics, constructs... A model predictive control framework incorporating a predictive model is constructed. This framework is used to optimize the traction or electric braking force commands and air braking force commands of the master and slave locomotives, outputting an optimal command sequence. An air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and a pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, corrects the traction or electric braking force command based on the execution error, outputs a braking command or final traction force, and feeds the execution error back to the vehicle-to-vehicle stability optimization layer. The braking command includes the final air braking command and / or electric braking force command. The execution error is used to correct the model parameters of the predictive model.

[0006] Optionally, the vehicle-to-ground speed safety layer calculates the total demand force of the train based on the status data, the track data, the real-time gradient of the master locomotive, and the real-time gradient of the slave locomotive, using the train force balance and forward gradient prediction. It then allocates the total demand force differentially based on the gradient difference between the real-time gradients of the master and slave locomotives. The output allocation results include: determining the desired acceleration based on the deviation between the current speed and the target speed; determining the total resistance based on the total train mass, basic drag coefficient, current speed, weighted average gradient, and curve radius; determining the total demand force of the train based on the inertial force corresponding to the desired acceleration and the total resistance; determining the gradient difference based on the real-time gradients of the master and slave locomotives; determining the safety boundary force of the master locomotive and the safety boundary force of the slave locomotive based on the total demand force of the train, the locomotive quantity allocation coefficient, and the gradient difference; and determining the adjustment margin of the master and slave locomotives based on a first set ratio of the maximum traction force or a second set ratio of the braking force.

[0007] Optionally, the model predictive control framework is used to optimize the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive, and the optimal command sequence is output by: predicting the position of the master locomotive and the position of the slave locomotive after the delay based on the communication delay, the current speed and acceleration of the master locomotive and the slave locomotive; extracting the gradient corresponding to the position of the master locomotive and the position of the slave locomotive after the delay based on the line data, and determining the effective gradient difference based on the two; and determining the slave locomotive feedforward based on the effective gradient difference. The compensation amount is then added to the traction or braking force command of the slave locomotive. An asynchronous prediction model is established based on the update cycles of the master locomotive and the slave locomotive commands. The future state of the slave locomotive in the asynchronous prediction model is determined based on the current command of the slave locomotive and the historical lagging command of the master locomotive. An asynchronous synchronization deviation index is determined based on the current command of the master locomotive and the historical command of the slave locomotive. An asynchronous rate constraint is determined based on the change rate of the current command of the master locomotive and the change rate of the historical command of the slave locomotive.

[0008] Optionally, the optimal instruction sequence output by the vehicle-to-vehicle stability optimization layer includes: using the traction or electric braking force of the master locomotive, the traction or electric braking force of the slave locomotive, the air braking force of the master locomotive, and the air braking force of the slave locomotive as decision variables; constructing a multi-objective cost function, which includes a term for the square of the coupler force at the slave locomotive, a penalty term for the rate of change of the traction or electric braking force of the master locomotive and the slave locomotive, a penalty term for the amount of air braking input of the master locomotive and the slave locomotive, and a penalty term for asynchronous synchronization deviation; solving the decision variables based on the train longitudinal dynamics equation, the physical limits of the actuators, the safety threshold of the coupler force at the slave locomotive, the speed safety boundary constraints determined by the safety boundary forces of the master locomotive, the safety boundary forces of the slave locomotive, the adjustment margins of the master locomotive and the slave locomotive, the asynchronous coordination constraints corresponding to the asynchronous synchronization deviation index, and the asynchronous rate constraints to obtain the optimal instruction sequence.

[0009] Optionally, after outputting the optimal instruction sequence, the method further includes: incorporating the multi-objective cost function and constraints into the model predictive control framework, solving it in a rolling manner within each control cycle, and outputting the optimal instruction sequences corresponding to the master locomotive and the slave locomotive respectively; subjecting the optimal instruction sequence to rate limiting and smoothing based on the force change rate limit and smoothing time constant; wherein the master locomotive and the slave locomotive have different smoothing time constants and issuance time offsets; and recording the asynchronous deviation of the current cycle and the coupler force response at the slave locomotive for parameter adjustment of the predictive model in the next optimization cycle.

[0010] Optionally, the air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, and corrects the traction force or electric braking force command based on the execution error. This includes: establishing a positive mapping between brake cylinder pressure and air braking force and the pre-trained air braking inverse model based on air braking system test data; determining the nominal decompression amount based on the air braking force command in the optimal command sequence and the pre-trained air braking inverse model; determining the actual air braking force based on the actual brake cylinder pressure measurement value and the positive mapping; determining the air braking force execution error based on the air braking force command and the actual air braking force; determining the traction force or electric braking force compensation amount based on the air braking force execution error; and superimposing the traction force or electric braking force compensation amount onto the traction force or electric braking force command in the optimal command sequence to obtain the final traction force or electric braking force command.

[0011] Optionally, outputting braking commands or final traction force and feeding back the execution error to the vehicle-to-vehicle stability optimization layer includes: determining the actual total force of the master locomotive and the actual total force of the slave locomotive based on the final traction force or electric braking force command and the actual air braking force; determining the steady-state tracking error based on the sum of the traction force or electric braking force command and the air braking force command, the actual total force of the master locomotive and the actual total force of the slave locomotive; when the steady-state tracking error meets a preset threshold, outputting the actual total force of the master locomotive and the actual total force of the slave locomotive to the train actuator; and feeding back the air braking force execution error and the actual coupler force to the vehicle-to-vehicle stability optimization layer.

[0012] This application also provides a multi-level cooperative control system for heavy-haul combined trains, including: The data acquisition module is used to acquire train status data, track data, real-time gradient of the main control locomotive, and real-time gradient of the slave control locomotive. The vehicle-to-ground speed safety layer module is used to calculate the total demand force of the train based on status data, track data, real-time gradient of the master locomotive and the slave locomotive, and prediction of the forward gradient based on the force balance of the train. It then distributes the total demand force of the train differentially based on the gradient difference between the master locomotive and the slave locomotive, and outputs the distribution results. The distribution results include the safety boundary force of the master locomotive, the safety boundary force of the slave locomotive, the adjustment margin of the master locomotive, and the adjustment margin of the slave locomotive.

[0013] The vehicle-to-vehicle stability optimization layer module receives the allocation results and, with the goal of minimizing the coupler force at the slave locomotive and constrained by communication delay and power asynchrony characteristics, constructs a model predictive control framework containing a predictive model. The model predictive control framework is used to optimize the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive, and outputs the optimal command sequence. The air-electric robust execution layer module is used to receive the optimal command sequence, determine the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determine the actual air braking force and execution error based on the actual brake cylinder pressure, correct the traction force or electric braking force command based on the execution error, output the braking command or final traction force, and feed the execution error back to the vehicle stability optimization layer; the braking command includes the final air braking command and / or electric braking force command; the execution error is used to correct the model parameters of the prediction model.

[0014] Optionally, the vehicle-to-ground speed safety layer module includes: The desired acceleration determination unit can be used to determine the desired acceleration based on the deviation between the current velocity and the target velocity; The total resistance determination unit can be used to determine the total resistance based on the total train mass, basic resistance coefficient, current speed, weighted average gradient, and curve radius. The total demand force determination unit can be used to determine the total demand force of a train based on the inertial force and total resistance corresponding to the desired acceleration. The gradient difference determination unit can be used to determine the gradient difference based on the real-time gradient of the master locomotive and the real-time gradient of the slave locomotive; The safety boundary force determination unit can be used to determine the safety boundary force of the master locomotive and the safety boundary force of the slave locomotive based on the total demand force of the train, the locomotive quantity allocation coefficient and the gradient difference; The adjustment margin determination unit can be used to determine the adjustment margin of the master locomotive and the slave locomotive based on a first set ratio of the maximum traction force or a second set ratio of the braking force.

[0015] Optionally, the vehicle-to-vehicle stability optimization layer module further includes: The rolling solver unit can be used to incorporate multi-objective cost functions and constraints into the model predictive control framework, and perform rolling solver in each control cycle to output the optimal instruction sequence for the master locomotive and the slave locomotive respectively. The smoothing delivery unit can be used to rate limit and smooth the delivery of the optimal command sequence based on the force change rate limit and the smoothing time constant; wherein the master locomotive and the slave locomotive have different smoothing time constants and delivery time offsets; The recording feedback unit can be used to record the asynchronous deviation of the current cycle and the coupler force response at the slave locomotive, so as to be used for parameter adjustment of the prediction model in the next optimization cycle.

[0016] This application acquires train status data, track data, and real-time gradients of both the master and slave locomotives. This allows the vehicle-to-ground speed safety layer to calculate the total required force of the train based on the train's force balance and the predicted gradient ahead. The force is then allocated differentially according to the gradient difference between the master and slave locomotives, thus resolving the force distribution mismatch problem caused by average gradient calculations. The vehicle-to-vehicle stability optimization layer aims to minimize the coupler force at the slave locomotive. It uses communication delay and power asynchrony characteristics as constraints to construct a model predictive control framework including a predictive model, optimizing traction or electric braking force commands and air braking force commands to suppress longitudinal impulses. The air-electric robust execution layer determines the nominal decompression based on the air braking force command and a pre-trained air brake inverse model. It then determines the actual air braking force and execution error based on the actual brake cylinder pressure. The execution error is used to correct the traction or electric braking force command, and the error is fed back to the vehicle-to-vehicle stability optimization layer, improving the tracking accuracy of air braking under nonlinear, time-delay, and hysteresis characteristics. This application achieves coordinated processing of speed safety, longitudinal stability and execution robustness at different levels, which can significantly improve the safety, stability and reliability of complex ramp operation.

[0017] This application also provides a multi-level cooperative control system for heavy-load combined trains, which has the above-mentioned beneficial effects, and will not be elaborated here. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a multi-level cooperative control method for heavy-haul combined trains provided in this application embodiment; Figure 2 This is a schematic diagram of a three-layer collaborative control architecture for heavy-haul combined trains provided in an embodiment of this application; Figure 3 This is a schematic diagram of a multi-level cooperative control system for a heavy-duty combined train, provided as an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, 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.

[0021] See Figure 1 , Figure 1 A flowchart illustrating a multi-level cooperative control method for heavy-haul combined trains provided in this application embodiment. This application can be executed via a train operation control system, which may include a vehicle-to-ground speed safety layer, a vehicle-to-vehicle smoothness optimization layer, and a pneumatic-electric robust execution layer. The method includes: S101, acquires train status data, track data, real-time gradient of the main control locomotive and the real-time gradient of the slave control locomotive; S102, the vehicle-to-ground speed safety layer calculates the total demand force of the train based on the status data, the track data, the real-time gradient of the master locomotive, and the real-time gradient of the slave locomotive, based on the train force balance and the predicted forward gradient. It then performs differential allocation of the total demand force of the train based on the gradient difference between the real-time gradients of the master locomotive and the slave locomotive, and outputs the allocation result. The allocation result includes the safety boundary force of the master locomotive, the safety boundary force of the slave locomotive, the adjustment margin of the master locomotive, and the adjustment margin of the slave locomotive. S103, the vehicle-to-vehicle stability optimization layer receives the allocation result, aims to minimize the coupler force at the slave locomotive, and constructs a model predictive control framework containing a predictive model, constrained by communication delay and power asynchrony characteristics; it then uses the model predictive control framework to optimize the traction force or electric braking force command and air braking force command of the master locomotive and slave locomotive, and outputs the optimal command sequence. S104, the air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, corrects the traction force or electric braking force command based on the execution error, outputs the braking command or final traction force, and feeds the execution error back to the vehicle-to-vehicle stability optimization layer; the braking command includes the final air braking command and / or electric braking force command; the execution error is used to correct the model parameters of the prediction model.

[0022] In step S101, train status data, track data, real-time gradient of the master locomotive, and real-time gradient of the slave locomotive can be acquired. Train status data may include at least one of the following: current train speed, target speed, total train mass, position, speed, acceleration, actual output value of traction or electric braking force, air brake cylinder pressure, and coupler force. Track data may include at least one of the following: forward gradient sequence, curve radius, track speed limit, phase separation zone position, and signal clearance information. The real-time gradient of the master locomotive and the real-time gradient of the slave locomotive can respectively represent the track gradient at the current position of the master locomotive and the current position of the slave locomotive. Real-time gradients can be obtained through satellite positioning, odometer, track database matching, inertial measurement unit, etc. Train status data and track data can be acquired through at least one of the following: onboard sensors, wireless communication, ground control center, and LKI track information database. The acquired data can provide the input basis for subsequent vehicle-to-ground speed safety layer, vehicle-to-vehicle smoothing optimization layer, and air-electric robust execution layer. By simultaneously acquiring the real-time gradient of both the master locomotive and the slave locomotive, force distribution deviations caused by using only the average gradient of the entire train can be avoided.

[0023] In step S102, the vehicle-to-ground speed safety layer can calculate the total demand force of the train based on the status data, track data, real-time gradient of the master locomotive, and real-time gradient of the slave locomotive, based on the train's force balance and forward gradient prediction. It then allocates the total demand force differentially based on the gradient difference between the master and slave locomotives, outputting the allocation result. The allocation result may include the master locomotive's safety boundary force, the slave locomotive's safety boundary force, the master locomotive's adjustment margin, and the slave locomotive's adjustment margin. The vehicle-to-ground speed safety layer aims to ensure that the train speed strictly follows the safe operating speed curve, controlling speed deviation within a safe threshold. The total demand force of the train is used to ensure safe train operation speed and can be the longitudinal force of the entire train required to meet the target speed; positive values ​​represent traction demand, and negative values ​​represent braking demand. The calculation of the total demand force based on the train's force balance and forward gradient prediction can consider at least one of the following: basic resistance, gradient-additional resistance, curve-additional resistance, and inertial force. Differential allocation based on gradient differences allows the master and slave locomotives to obtain more matched traction or braking forces at different gradients. The safety boundary forces of the master and slave locomotives can serve as the speed safety boundaries for the car-to-car stability optimization layer. The adjustment margins of the master and slave locomotives represent the allowable adjustment range of the car-to-car stability optimization layer near the safety boundaries, thus ensuring that the overall speed safety constraints are not exceeded during stability optimization. The car-to-ground speed safety layer can output the allocation results to the car-to-car stability optimization layer.

[0024] In one feasible implementation, this step may include the following sub-steps: The first step is to determine the desired acceleration based on the deviation between the current speed and the target speed.

[0025] The deviation between the current speed and the target speed represents the difference between the train's current operating speed and the desired operating speed. The desired acceleration can be determined by taking the negative of the product of the speed deviation gain and the speed deviation. The speed deviation gain can be tuned according to the train's response characteristics, with a typical range of 0.1 to 0.3 per second. If the current speed is lower than the target speed, the desired acceleration can be positive, indicating that acceleration is needed; if the current speed is higher than the target speed, the desired acceleration can be negative, indicating that deceleration is needed. By setting the speed deviation gain, the speed tracking speed and smoothness can be adjusted.

[0026] The second step is to determine the total resistance based on the total train mass, basic resistance coefficient, current speed, weighted average gradient, and curve radius.

[0027] Total resistance can include basic resistance, gradient resistance, and curve resistance. Basic resistance can be determined based on the total train mass, basic resistance coefficient, and current speed. The basic resistance coefficient can include constant, linear, and quadratic coefficients, which can be selected from a standard table based on the vehicle type. For example, for a C80 freight car, the constant coefficient can be 1.65 N / kN. The coefficient of the linear term can be taken as 0.0247 N / kN / km / h. The coefficient of the quadratic term can be taken as 0.000428 N / kN / km / h squared (i.e., The additional resistance due to gradients can be determined based on the total train mass, gravitational acceleration, and weighted average gradient. The weighted average gradient represents the average gradient over the train's length; it can be positive for uphill sections and negative for downhill sections. Gravitational acceleration can be taken as 9.81 m / s². The additional resistance due to curves can be determined based on the curve radius converted to unit gradient resistance. The curve radius can be the average curve radius at the train's location, and the unit can be meters. By combining the basic resistance, the additional resistance due to gradients, and the additional resistance due to curves, the total resistance of the train can be obtained.

[0028] The third step is to determine the total required force of the train based on the inertial force and total resistance corresponding to the desired acceleration.

[0029] The inertial force corresponding to the desired acceleration can be determined by multiplying the total train mass by the desired acceleration. Superimposing the inertial force with the total resistance yields the total traction or braking force required for the entire train. This total required force can serve as the baseline force for subsequent allocation calculations, representing the total longitudinal force required by the entire train to meet the speed target. A positive total required force indicates traction demand; a negative value indicates braking demand.

[0030] The fourth step is to determine the gradient difference based on the real-time gradient of the master locomotive and the slave locomotive.

[0031] The gradient difference represents the degree of difference between the gradient of the master locomotive and the gradient of the slave locomotive. The gradient difference can be positive or negative. This difference reflects the difference in gradient between the front and rear sections of the train and is the core basis for differentiated force distribution. For example, if the master locomotive is going uphill while the slave locomotive is going downhill, the gradient difference can be large. The master locomotive may require more traction or less braking force, while the slave locomotive may require less traction or more braking force.

[0032] The fifth step is to determine the safety boundary force of the master locomotive and the safety boundary force of the slave locomotive based on the total demand force of the train, the locomotive quantity allocation coefficient, and the gradient difference.

[0033] The locomotive allocation coefficient can be set according to the locomotive traction capacity ratio, ensuring that the sum of the allocation coefficients for the master locomotive and the slave locomotive is 1. For example, when there is one master locomotive and one slave locomotive, both allocation coefficients can be set to 0.5. If the number of locomotives is different, they can be weighted by power or axle count. The gradient compensation coefficient represents the force required to compensate for a unit gradient difference. It can be determined by the train's unit gradient resistance, and a typical value is the product of the total train mass and gravitational acceleration divided by 1000, expressed in kilonewtons per percentage. The master locomotive safety boundary force can be determined by multiplying the master locomotive allocation coefficient by the total train demand force, and then subtracting the product of the gradient compensation coefficient and the gradient difference. The slave locomotive safety boundary force can also be determined by multiplying the slave locomotive allocation coefficient by the total train demand force, and then subtracting the product of the gradient compensation coefficient and the gradient difference. By introducing the gradient difference, the safety boundary forces of the master and slave locomotives can better reflect the actual stress state on a gradient.

[0034] The sixth step is to determine the adjustment margin of the master locomotive and the adjustment margin of the slave locomotive based on the first set ratio of the maximum traction force or the second set ratio of the braking force.

[0035] The first and second setting ratios can be the same or different. Typically, they can be set to 5% to 10% of the maximum traction or braking force. The adjustment margin serves as the range within which the vehicle-to-vehicle stability optimization layer can fluctuate near the safety boundary during stability optimization. By setting the adjustment margin, it can be ensured that the optimization adjustment of the vehicle-to-vehicle stability optimization layer does not exceed the overall speed safety constraints, while retaining necessary stability optimization space.

[0036] In step S103, the vehicle-to-vehicle stability optimization layer receives the allocation results and constructs a model predictive control framework (MMC) containing a predictive model, with the objective of minimizing the coupler force at the slave locomotive and constraints of communication delay and power asynchrony characteristics. The MMC optimizes the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive using the MMC, outputting the optimal command sequence. The vehicle-to-vehicle stability optimization layer can use the traction or electric braking force of the master locomotive, the traction or electric braking force of the slave locomotive, the air braking force of the master locomotive, and the air braking force of the slave locomotive as decision variables. The MMC can solve the problem in a rolling fashion within each control cycle, outputting the optimal command sequence for each locomotive. Communication delay can represent the inherent transmission delay of the wireless communication system, typically ranging from 0.5 seconds to 1.0 seconds. Power asynchrony characteristics represent the asynchrony in command updates and responses between the master locomotive and slave locomotive due to communication delay and independent clock scheduling. By using communication latency and dynamic asynchrony characteristics as constraints, the distributed control features of heavy-haul combined trains can be more realistically reflected. Minimizing the coupler force at the slave locomotive directly suppresses longitudinal impulses and improves operational stability. Optimizing the air braking force commands enables differentiated allocation of air braking between the master and slave locomotives, thereby mitigating longitudinal surges caused by asynchrony. The optimal command sequence output by the car-to-car stability optimization layer can be passed to the air-electric robust execution layer.

[0037] In one feasible implementation, optimizing the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive using a model predictive control framework, and outputting the optimal command sequence may include the following steps: The first step is to predict the position of the master locomotive and the position of the slave locomotive after the delay based on the communication delay, the current speed and acceleration of the master locomotive and the slave locomotive.

[0038] Based on the current motion state, the position after the time delay can be predicted according to the uniform acceleration relationship. For example, the position after the time delay can be determined by the sum of the current position, the product of the current velocity and the time delay, and the product of half of the current acceleration and the square of the time delay. By predicting the position after the time delay, the track section where the master locomotive and the slave locomotive will be located in the future can be known in advance.

[0039] The second step is to extract the gradient corresponding to the position of the master locomotive and the gradient corresponding to the position of the slave locomotive after the time delay based on the line data, and determine the effective gradient difference based on the two.

[0040] The effective gradient difference represents the actual difference between the gradient of the master locomotive after the time delay and the gradient of the slave locomotive after the time delay. The effective gradient difference can compensate for the resistance imbalance caused by gradient changes during the communication delay period.

[0041] The third step is to determine the feedforward compensation amount of the slave locomotive based on the effective gradient difference, and then add the feedforward compensation amount of the slave locomotive to the traction or braking force command of the slave locomotive.

[0042] The feedforward compensation amount can be determined by multiplying the compensation coefficient by the effective gradient difference. The compensation coefficient can be the same as the gradient compensation coefficient in the vehicle-to-ground speed safety layer. This feedforward compensation amount can be directly superimposed on the traction or braking force command of the slave locomotive to preemptively offset the resistance imbalance caused by gradient changes during the time delay period.

[0043] The fourth step is to establish an asynchronous prediction model in the prediction model based on the update cycle of the master locomotive command and the update cycle of the slave locomotive command.

[0044] In the asynchronous predictive model, the future state of the slave locomotive is determined based on its current commands and the historical, lagging commands from the master locomotive. The update cycles for the master and slave locomotive commands can differ. Within the prediction time domain of model predictive control, the actual state response of the slave locomotive depends not only on its current commands but also on the lagging influence of the master locomotive's historical commands. Therefore, the predicted state of the slave locomotive at a future moment can be determined based on its current state, the master locomotive's historical, lagging commands, and the slave locomotive's current traction or electric braking force and air braking force to be optimized. By establishing an asynchronous predictive model, the future response of the slave locomotive can be predicted more accurately.

[0045] The fifth step is to determine the asynchronous synchronization deviation index based on the current instructions of the master locomotive and the historical instructions of the slave locomotive.

[0046] The asynchronous-synchronization deviation index represents the difference between the current command of the master locomotive and the historical command of the slave locomotive, quantifying the degree of timing asynchrony between the master and slave commands. A larger index indicates a more significant deviation between the current master command and the slave's historical command.

[0047] The sixth step is to determine the asynchronous rate constraint based on the current command change rate of the master locomotive and the historical command change rate of the slave locomotive. The asynchronous rate constraint indicates that the absolute value of the difference between the current command change rate of the master locomotive and the historical command change rate of the slave locomotive must not exceed the allowable asynchronous rate difference limit. A typical allowable asynchronous rate difference limit is 10 kN / s. This constraint forces the sudden change rates of master and slave commands to remain coordinated, preventing one side from changing drastically while the other responds with a lag.

[0048] When outputting the optimal instruction sequence, the following steps may be included: The traction force or electric braking force of the master locomotive, the traction force or electric braking force of the slave locomotive, the air braking force of the master locomotive, and the air braking force of the slave locomotive are used as decision variables. Construct a multi-objective cost function, which includes a term for the square of the coupler force at the slave locomotive, a penalty term for the rate of change of traction force or electric braking force of the master locomotive and the slave locomotive, a penalty term for the amount of air brake input of the master locomotive and the slave locomotive, and a penalty term for asynchronous synchronization deviation. Based on the train longitudinal dynamics equation, the physical limits of the actuator, the safety threshold of the coupler force at the slave locomotive, the speed safety boundary constraints determined by the safety boundary forces of the master locomotive, the safety boundary forces of the slave locomotive, the adjustment margins of the master locomotive and the slave locomotive, the asynchronous coordination constraints corresponding to the asynchronous-synchronous deviation index, and the asynchronous rate constraints, the decision variables are solved to obtain the optimal command sequence.

[0049] Traction force or electric braking force can be positive to represent traction and negative to represent electric braking. Air braking force can be non-positive, with 0 indicating no air braking is applied. By incorporating air braking force into the decision variable, differentiated optimization of air braking in master-slave locomotives can be achieved.

[0050] The square term for the coupler force at the locomotive control station can be used to directly suppress coupler force impact. The penalty term for the rate of change of traction or electric braking force can be used to suppress frequent fluctuations in commands. The penalty term for air brake deployment can be used to guide the priority use of electric braking and reduce air brake wear. The penalty term for asynchronous synchronization deviation can be used to directly penalize the degree of timing mismatch between master and slave control commands. Each item can correspond to a non-negative weight coefficient. The weighting principle can be that the weight of the coupler force suppression term is much greater than the weight of the rate of change penalty term, the air brake deployment penalty term can be taken with a large positive value to suppress unnecessary air brake deployment, and the asynchronous synchronization deviation penalty term can be taken with a moderate positive value to balance asynchronous coordination with the main objective. For example, the weight of the square term for coupler force can be 100, the weight of the penalty term for the rate of change of traction or electric braking force can be 1, the weight of the penalty term for air brake deployment can be 10, and the weight of the asynchronous synchronization deviation penalty term can be 5. The weights can be set according to implementation experience, or adjusted according to track conditions, train formation, and air brake system characteristics.

[0051] The longitudinal dynamics equations of the train describe the force balance relationship of each car, including locomotives and freight cars, encompassing inertial forces, coupler forces, basic resistance, and additional resistance from gradients and curves. The physical limits of the actuators can be determined by the traction characteristic curves, electric braking characteristic curves, and maximum decompression of the air braking system for each locomotive. The safety threshold for the coupler force at the slave locomotive can be set based on the strength of the coupler and buffer, typically 1500 kN. The speed safety boundary constraint represents the total output force of this layer, i.e., the sum of traction force or electric braking force and air braking force, which must be adjusted within a limited range near the safety boundary to ensure that the smooth optimization of the car-to-car smoothness optimization layer does not compromise the speed safety of the car-to-ground layer. The asynchronous coordination constraint represents the absolute value of the asynchronous synchronization deviation index not exceeding the allowable threshold, typically 20 kN. The asynchronous rate constraint can be added as a hard constraint to the optimization framework. By solving the decision variables, the optimal command sequence can be obtained.

[0052] In step S104, the air-electric robust execution layer can receive the optimal command sequence, determine the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determine the actual air braking force and execution error based on the actual brake cylinder pressure, correct the traction force or electric braking force command based on the execution error, output the braking command or final traction force, and feed the execution error back to the vehicle stability optimization layer. The braking command includes the final air braking command and / or electric braking force command, and the execution error is used to correct the model parameters of the prediction model.

[0053] Execution error can be used to correct the model parameters of the prediction model. The air-electric robust execution layer can design dynamic coordination and robust compensation strategies for traction or electric braking force and air braking force on both the master and slave locomotives, addressing the nonlinearity, time delay, and hysteresis characteristics of the air braking system. The air braking inverse model can be pre-trained based on air braking system test data. The nominal decompression can be expressed as the theoretical decompression required to achieve the target air braking force. The actual brake cylinder pressure can be measured in real time by a pressure sensor. The actual air braking force can be determined by the positive mapping between brake cylinder pressure and air braking force. Execution error can represent the deviation between the target air braking force and the actual air braking force. Correcting the traction or electric braking force command based on the execution error allows for leveraging the fast response speed of the traction motor to increase electric braking force when air braking force is insufficient and decrease it when air braking force overshoots. In this step, the final braking command or final traction force is output to the train actuator. Feeding the execution error back to the car-to-car stability optimization layer forms a closed-loop hierarchical control.

[0054] In one feasible implementation, the air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, and corrects the traction force or electric braking force command based on the execution error. This can be achieved in the following manner: A forward mapping between brake cylinder pressure and air braking force is established based on air braking system test data, along with the pre-trained air braking inverse model. The nominal pressure reduction is determined based on the air braking force command in the optimal command sequence and the pre-trained air braking inverse model. The actual air braking force is determined based on the actual brake cylinder pressure measurement and the forward mapping. The air braking force execution error is determined based on the air braking force command and the actual air braking force. The traction force or electric braking force compensation is determined based on the air braking force execution error. The traction force or electric braking force compensation is superimposed on the traction force or electric braking force command in the optimal command sequence to obtain the final traction force or electric braking force command.

[0055] Air braking systems can exhibit nonlinear characteristics such as hysteresis, dead zone, and pressure rise rate limitation. A mapping relationship between brake cylinder pressure and air braking force can be established based on bench or circuit test data of the air braking system. A forward mapping can determine the actual air braking force based on brake cylinder pressure, the friction coefficient varying with pressure, the brake cylinder piston area, and the number of brake cylinders. The actual air braking force can be negative, representing the braking force. Based on the forward model, an inverse model can be further constructed to calculate the required nominal decompression from the target air braking force issued by the vehicle-to-vehicle layer. Calculating the nominal decompression using the inverse model allows for a close approximation of the target air braking force even when the air braking system exhibits nonlinear characteristics. The actual brake cylinder pressure can be measured by a pressure sensor. Execution error represents the deviation between the target command and the actual execution; a positive value indicates insufficient air braking force, while a negative value indicates overshoot.

[0056] The proportional-integral (PI) control law can include proportional gain and integral gain. The proportional gain can be tuned according to the system identification results, typically ranging from 0.5 to 1.0. The integral gain can also be tuned according to the system identification results, typically ranging from 0.05 to 0.2 seconds per second. This compensation mechanism can automatically adapt to the physical direction: when the execution error is greater than 0 (i.e., insufficient air braking force), the compensation term can be negative, increasing the final traction force or electric braking force in the negative direction to compensate for the shortfall; when the execution error is less than 0 (i.e., air braking force overshoot), the compensation term can be positive, adjusting the final traction force or electric braking force in the positive direction to restore balance. The master locomotive and slave locomotives can each execute local compensation independently. Through local closed-loop compensation, robust tracking of the actual output force to the vehicle-to-vehicle stability optimization layer's optimization commands can be ensured.

[0057] Accordingly, during output, the actual total force of the master locomotive and the actual total force of the slave locomotive can be determined based on the final traction force or electric braking force command and the actual air braking force; then, the steady-state tracking error can be determined based on the sum of the traction force or electric braking force command and the air braking force command, the actual total force of the master locomotive and the actual total force of the slave locomotive; then, when the steady-state tracking error meets a preset threshold, the actual total force of the master locomotive and the actual total force of the slave locomotive are output to the train actuator; finally, the air braking force execution error and the actual coupler force are fed back to the car-to-car stability optimization layer.

[0058] The actual total force of the master-controlled locomotive can be the sum of the compensated electric braking force and the actual air braking force. The actual total force of the slave-controlled locomotive can be the sum of the compensated electric braking force and the actual air braking force.

[0059] Steady-state tracking error represents the deviation between the actual total force and the optimized command total force. The permissible steady-state tracking error threshold can be 1% to 2% of the maximum actuated force. This threshold can be set by those skilled in the art based on implementation experience, or according to the actuator accuracy and track conditions. By setting a preset threshold, robust tracking of the actual output force to the second-level optimized command can be ensured. The vehicle-to-vehicle smoothing optimization layer can adjust weight coefficients or correct prediction model parameters in the next optimization cycle. Through feedback, closed-loop hierarchical control can be formed, improving the safety and reliability of long-term operation.

[0060] See Figure 2 , Figure 2 This is a schematic diagram of a three-layer cooperative control architecture for heavy-load combined trains provided in an embodiment of this application. Figure 2 As shown, the left side inputs include train status data, track data, real-time gradient of the master locomotive, and real-time gradient of the slave locomotive. The upper vehicle-to-ground cooperative speed safety control layer calculates the total traction / braking force demand of the entire train and distributes traction / braking force based on gradient differences, outputting traction / braking force distribution for the master and slave locomotives that meets speed safety requirements. The middle vehicle-to-vehicle cooperative smooth optimization control layer performs time delay compensation, constructs a model predictive control framework, and outputs the optimal sequence commands to obtain traction / electric braking and air braking commands for the master and slave locomotives that meet smoothness requirements. The lower air-electric cooperative robust execution control is executed for both slave and master locomotives, outputting traction / electric braking and air braking commands for both the master and slave locomotives to the corresponding locomotives, and collecting the status of the slave and master locomotives to feed back to the upper layer, forming a hierarchical closed-loop control. The bottom shows a heavy-haul combined train formation consisting of slave locomotives, freight cars, and master locomotives.

[0061] In another feasible implementation, after outputting the optimal instruction sequence, the following steps may also be included: The multi-objective cost function and constraints are incorporated into the model predictive control framework, and the solution is applied on a rolling basis within each control cycle to output the optimal command sequence for both the master locomotive and the slave locomotive. The optimal command sequence is rate-limited and smoothly issued based on the force change rate limit and the smoothing time constant. The master locomotive and the slave locomotive can have different smoothing time constants and issuance time offsets.

[0062] For example, the limit for the rate of change of traction force or electric braking force can be set to 50 kN / s, and the limit for the rate of change of air braking force can be set to 20 kN / s. The smoothing time constant of the master locomotive can be set to 2.0 seconds, and the smoothing time constant of the slave locomotive can be set to 1.5 seconds to reflect the asynchronous smoothing characteristics. When smoothing the output, a smoothing output law can be used, which may include: dividing the difference between the optimal target command and the current actual output force by the smoothing time constant, and then multiplying by the control cycle to obtain the expected increment before the limit; limiting the expected increment before the limit according to the force change rate limit; and adding the increment after the limit to the current actual output force. By independently setting different smoothing time constants and output time offsets for the master and slave locomotives, staggered output can be achieved, further weakening the synchronous impact. Record the asynchronous deviation of the current cycle and the coupler force response at the slave locomotive for parameter adjustment of the prediction model in the next optimization cycle. By rolling solution, smoothing output, and recording feedback, the real-time performance and stability of control can be improved.

[0063] The parameters in the above embodiments of this application can be set based on implementation experience. For example, the speed deviation gain can be set to 0.1 to 0.3 per second based on the train response characteristics. The basic drag coefficient can be selected from a standard table based on the vehicle type. The adjustment margin can be set to 5% to 10% of the maximum traction force or maximum braking force. The communication delay can be measured based on the wireless communication system, typically 0.5 to 1.0 seconds. The coupler force safety threshold can be set to 1500 kN based on the coupler and buffer strength. The asynchronous synchronization deviation threshold can be set to 20 kN. The asynchronous rate of change difference limit can be set to 10 kN per second. The traction force or electric braking force rate of change limit can be set to 50 kN per second. The air braking force rate of change limit can be set to 20 kN per second. The master locomotive smoothing time constant can be set to 2.0 seconds. The slave locomotive smoothing time constant can be set to 1.5 seconds. The proportional gain can be set to 0.5 to 1.0. The integral gain can be set to 0.05 to 0.2 per second. The steady-state tracking error threshold can be set to 1% to 2% of the maximum actuation force. The above values ​​are for illustrative purposes only. Specific values ​​can be set by those skilled in the art based on implementation experience, track conditions, train formation, and air braking system characteristics.

[0064] It should be noted that the vehicle-to-ground speed safety layer, vehicle-to-vehicle stability optimization layer, and air-electric robust execution layer can each correspond to an independent control module, or they can be integrated into the same control chip. Data can be transmitted between the layers via onboard network, wireless communication link, or ground communication link. The model predictive control framework can adopt a rolling time-domain optimization approach, with each control cycle including data acquisition, state prediction, optimization solution, smoothing deployment, and execution feedback. Through hierarchical fusion and closed-loop feedback, the operational safety, stability, and control reliability of heavy-haul combined trains under complex operating conditions can be significantly improved.

[0065] For vehicle-to-ground speed safety layer: Inputs include: real-time train speed Target speed Total mass of the train Line data such as the gradient sequence and curve radius of the LKJ line ahead, and the real-time gradient of the main control locomotive's current position. Real-time gradient of the current position of the slave locomotive .

[0066] The output allocation results include: the total force demand of the entire train allocated to the traction / braking forces of the master locomotive and slave locomotives. , and the main control locomotive adjustment margin Locomotive adjustment margin , which serves as the speed safety boundary constraint for the vehicle-to-vehicle stability optimization layer.

[0067] In each control cycle, firstly based on the current speed of the train... relative to target speed The deviation is used to calculate the desired acceleration. The velocity deviation gain is defined as... (Set according to train response characteristics, typical value range is 0.1~0.3s) - ¹), then the expected acceleration The calculation formula is as follows:

[0068] Then, the total resistance experienced by the entire train during operation is calculated, including basic resistance, gradient resistance, and curve resistance. The sum of basic resistance and gradient resistance is denoted as... Its calculation uses the standard resistance formula. In the formula, The total mass of the entire train The current train speed, The acceleration due to gravity is taken as 9.81 m / s². This is the weighted average gradient within the train formation length (positive for uphill and negative for downhill). , , These are all basic drag coefficients, selected from the standard table according to the vehicle type (e.g., C80 truck). , , :

[0069] Curve Additional Resistance Calculated based on unit slope resistance, in the formula... The average curve radius (in meters) at the train's location:

[0070] The total force required by the train is obtained by superimposing the inertial force corresponding to the desired acceleration with the aforementioned resistance forces. :

[0071] This value serves as the baseline force for subsequent allocation calculations and represents the total longitudinal force required by the entire train to meet the speed target (positive values ​​indicate traction, and negative values ​​indicate braking).

[0072] Obtain the gradient of the main control locomotive and the gradient of the controlled locomotive Calculate the slope difference :

[0073] The gradient difference reflects the degree of difference in the slope between the front and rear parts of the train, and is the core basis for differentiated force distribution.

[0074] The force distribution coefficients for the master locomotive and the slave locomotive are defined as follows: and This coefficient is set according to the locomotive's traction capacity ratio to meet the following requirements. For example, when there is one locomotive and one slave locomotive. If the number of locomotives differs, weighting is applied based on power or number of axles. Gradient compensation coefficient. The physical meaning of is the force required to compensate for a unit gradient difference, determined by the train's unit gradient resistance, typically taken as . (unit This ensures accurate compensation for the difference in gravity components caused by the gradient difference. Based on the above parameters, the traction / braking force distribution between the master locomotive and the slave locomotive is as follows:

[0075] Finally, the output of the main control locomotive's traction / braking force. With the traction / braking force of the slave locomotive As the speed safety boundary of the car layer, and setting the allowable adjustment margin. , (Usually taken as 5% to 10% of the maximum traction / braking force, unit) This is the range within which the vehicle-to-vehicle stability optimization layer is allowed to fluctuate around the safety boundary during stability optimization, ensuring that the optimization adjustment of the vehicle-to-vehicle stability optimization layer does not exceed the overall constraint of speed safety.

[0076] For the vehicle-to-vehicle stability optimization layer: Input: Velocity safety boundary force transmitted by the vehicle-to-ground velocity safety layer , , , Inherent delay in wireless communication (Based on actual system measurements, typical value 0.5–1.0 s); The current state vectors of the master locomotive and slave locomotives, including the position of the master locomotive. Locomotive position Main locomotive speed Locomotive speed control Main locomotive acceleration Slave locomotive acceleration Track gradient database; traction / electric braking characteristic curves of each locomotive and execution capability parameters such as maximum decompression of the air braking system.

[0077] Output: Optimal command sequence, including traction / electric braking force of the main locomotive. Main locomotive air braking system Traction / electric braking force of locomotive From the control of locomotive air braking force After rate smoothing, it is sent to the air-powered robust execution layer.

[0078] Wireless communication systems have inherent transmission delays This delay means that by the time the slave locomotive receives the master control command, the master locomotive has already moved to a new position ahead, and its actual force environment has changed. To offset this delay, the system first predicts based on the current motion state. The positions of the master and slave locomotives after the specified time. (Order) , Let the current positions of the master locomotive and the slave locomotive be respectively. Then, the predicted position is given by the formula for uniformly accelerated motion: ; ; The actual track gradients at the predicted locations are extracted from the track database and denoted as follows: and And define the effective slope difference :

[0079] Effective slope difference This reflects the actual difference in gradient between the master and slave locomotives after the time delay. Based on this, the feedforward compensation amount of the slave locomotive is defined. :

[0080] Where the compensation coefficient and With consistent values, this compensation term is directly superimposed on the traction / braking force command of the slave locomotive to offset the resistance imbalance caused by gradient changes during the time delay period.

[0081] To reflect the asynchronous power characteristics of the master-slave locomotives due to communication delays and independent clock scheduling, the instruction update cycle of the master locomotive is assumed to be... The instruction update cycle for the locomotive is ,allow In the prediction time domain of model predictive control, the actual state response of the slave locomotive depends not only on its current command but also on the lag effect of the master locomotive's historical commands. Therefore, the slave locomotive's future... Predicted state at time Revised to:

[0082] in Let the longitudinal dynamic state transition function of the slave locomotive be denoted as . This is the current state vector (including speed, position, etc.) of the slave locomotive. for Traction / electric braking commands issued by the main control locomotive before the scheduled time. and These are the traction / electric braking force and air braking force that the slave locomotive currently needs to optimize.

[0083] Furthermore, an asynchronous-synchronization deviation index is defined. This is used to quantify the degree of timing asynchrony between master and slave control instructions.

[0084] in for The value of the master control locomotive's own command before the current time. The larger this value is, the more significant the deviation between the master control command and the slave control historical command at the current time.

[0085] To suppress vertical surges caused by excessive differences in instruction rates, an asynchronous rate constraint is defined. Let... The rate of change of the current master instruction (i.e.) ), To control the rate of change of historical instructions, The allowable asynchronous rate of change difference limit (typically 10). Then the constraint is expressed as:

[0086] This constraint forces the master and slave control commands to maintain a coordinated mutation rate, preventing one side from changing drastically while the other side responds with a lag.

[0087] The decision variables for the asynchronous control model optimization problem include: the traction / electric braking force of the main locomotive. Traction / Electric Braking of Slave Locomotive (Positive values ​​indicate traction, negative values ​​indicate electric braking), and the main control locomotive's air braking force. From the control of locomotive air braking force (Values ​​are non-positive; 0 indicates no air braking is applied).

[0088] Set the prediction time domain as (Typical value 10s), the discretization step size is consistent with the control cycle. The primary objective is to minimize the coupler force impact at the slave locomotive, supplemented by penalties for traction / electric braking force change rate, air brake engagement amount, and asynchronous / synchronous deviation. The objective cost function is constructed as follows: :

[0089] in, This represents the longitudinal coupler force from the locomotive control point, and its squared term is used to directly suppress the impact of the coupler force. and These are the rates of change of the traction / electric braking force of the master locomotive and the rate of change of the traction / electric braking force of the slave locomotive at the current moment, respectively, used to suppress frequent fluctuations in commands; and These are the amplitudes of the air braking force command of the master locomotive and the amplitudes of the air braking force command of the slave locomotive at the current moment (taking non-positive numbers). The square of these values ​​is used as a penalty to guide the system to prioritize the use of electric braking and reduce air brake wear. The asynchronous-synchronization deviation index is defined, and its squared term is used to directly penalize the degree of timing mismatch between master and slave control instructions; to All are non-negative weight coefficients corresponding to each item. The weight allocation principle is as follows: (Ensure that coupler force suppression is dominant). and A larger positive value is used to suppress unnecessary activation of the air brakes. Choose a moderate positive value to balance asynchronous coordination with the primary objective. A typical value is... , , , .

[0090] The above decision variables must satisfy the following constraints: Constraint 1, the longitudinal dynamic equation of the train, describes the force balance relationship of each car (including locomotive and freight car), including inertial force, coupler force, basic resistance, and additional resistance on slopes and curves, as an equation constraint.

[0091] The second constraint, the physical upper and lower limits of the execution capability, are determined by the traction characteristic curves, electric braking characteristic curves, and maximum decompression of the air braking system of each locomotive, and serve as an inequality constraint.

[0092] Constraint 3: Coupler force safety threshold constraint. Let... The coupler force safety threshold (set according to the strength of the coupler and buffer, typical value 1500) Then, the coupler force at the locomotive control point must meet the following requirements:

[0093] Constraint 4: Speed ​​Safety Boundary Constraint. The safety boundary forces transmitted by the vehicle-to-ground speed safety layer are as follows: and The allowable adjustment margins are respectively and The total force output of this layer (the sum of traction / electric braking force and air braking force) must be adjusted within a limited range near the safety boundary to ensure that the smooth optimization of the vehicle-to-vehicle stability layer does not compromise the speed safety of the vehicle-to-ground layer.

[0094] Constraint 5: Asynchronous Coordination Constraint. This involves setting the asynchronous synchronization deviation index... Include explicit constraints, assuming their absolute value does not exceed the allowable threshold. (Typical value: 20) ):

[0095] Meanwhile, asynchronous rate constraints are also incorporated as hard constraints into the optimization framework.

[0096] The objective cost function and all constraints are integrated into the model predictive control for quadratic programming. The solution is obtained by rolling the calculation within each control cycle, synchronously outputting the optimal command sequence. , , , .

[0097] To avoid the impact of sudden instruction changes on the execution mechanism, the optimized instructions are smoothly issued after rate limiting. The actual output force at the current moment. The optimal target instruction. To control the cycle, Force change rate limit (traction / electric braking force taken as 50) Air braking force 20 ), For the smoothing time constant (taken from the main control locomotive) Take from the locomotive (To reflect the asynchronous smoothing characteristics). The smoothing output law is:

[0098] in As a saturation function, the expected increment before limiting is first filtered using a smoothing time constant, and then... The upper and lower limits are then superimposed onto the current output. This is the smoothing output law. Compatible with asynchronous control, different smoothing time constants and sending time offsets can be set independently for the master and slave locomotives to achieve staggered output and further reduce synchronous impact.

[0099] The optimized instructions obtained after the above steps of solution and smoothing are passed as expected inputs to the air-electric cooperative robust execution layer. Simultaneously, the asynchronous deviation of the current cycle is... and coupler force response Record and provide feedback for adaptive adjustment of model parameters in the next optimization cycle.

[0100] For the air-powered robust execution layer: Input: The optimal command issued by the vehicle-to-vehicle stability optimization layer. , (Independent inputs for master and slave controllers respectively); Real-time measured value of actual brake cylinder pressure ; Feedback value of actual output force of traction motor.

[0101] Output: The final traction / electric braking force command applied by the master and slave locomotives. , And the actual air braking commands issued to the braking system after inverse model conversion; and the final total force applied to the master and slave locomotives. , and execution error Feedback is sent to the vehicle-to-vehicle layer for parameter correction in the next optimization cycle.

[0102] Air braking systems exhibit nonlinear characteristics such as hysteresis, dead zone, and pressure rise rate limitation. To bring their nonlinear response under control, a mapping relationship between brake cylinder pressure and air braking force is first established based on bench or circuit test data of the air braking system. For brake cylinder pressure, The coefficient of friction varies with pressure. The piston area of ​​the brake cylinder. Given the number of brake cylinders, the actual air braking force is... (Negative values ​​represent braking force) are given by the following formula:

[0103] Based on this forward model, a pre-trained air braking inverse model is further constructed. The target air braking force issued by the car floor Calculate the required nominal pressure reduction It is then output to the air brake actuator.

[0104] Simultaneously, the actual brake cylinder pressure is collected in real time. Through actual air braking force The positive mapping is converted into actual aerodynamic braking force. Define the air braking force execution error. for:

[0105] This error reflects the deviation between the target command and the actual execution; a positive value indicates insufficient air braking force, and a negative value indicates overshoot.

[0106] During dynamic compensation of electric braking force, the traction / electric braking force output by the traction motor (symbols are uniformly defined as follows) (Positive for traction, negative for electric braking) The response speed is extremely fast (milliseconds). This advantage is utilized to construct a fast compensator based on air braking force execution error. In each local controller (executed independently by the master locomotive and slave locomotives), the final actual issued traction / electric braking force command... Original instructions issued from the car level Composition of superposition ratio-integral robust compensation term:

[0107] in For proportional gain, For integral gain, both are tuned according to the system identification results, with typical values ​​taken as follows: , The physical orientation of this compensation mechanism automatically adapts: when When there is insufficient air braking force, the compensation term is negative, making Increase in the negative direction (increase electric braking force to make up for the gap); when When there is an overshoot in the air braking force, the compensation term is positive, making... Adjust in the positive direction (reduce electric braking force to restore balance).

[0108] The total force actually output by the master locomotive and the slave locomotive is the sum of the compensated electric braking force and the actual air braking force, respectively.

[0109] in , It is calculated based on the actual brake cylinder pressure.

[0110] This layer uses local closed-loop compensation to ensure robust tracking of the actual output force to the optimized instructions of the second layer. The allowable steady-state tracking error threshold is defined as follows: (Generally, 1% to 2% of the maximum execution power is taken), then the steady-state tracking performance satisfies:

[0111] The actual total force output is sent to the train actuator, while the execution error of each cycle is also included. and actual coupler force Feedback is sent to the vehicle-to-vehicle stability optimization layer, which is used to adjust the weight coefficients or correct the prediction model parameters in the next optimization cycle to achieve hierarchical closed-loop control.

[0112] The train's longitudinal dynamics equations can consider the inertial force, coupler force, basic resistance, gradient resistance, and curve resistance of each car. The physical limits of the actuators can consider the traction characteristic curve, the electric braking characteristic curve, and the maximum pressure reduction of the air braking system. The forward mapping of the air brakes can consider the brake cylinder pressure, friction coefficient, brake cylinder piston area, and number of brake cylinders. The inverse model of the air brakes can be obtained by inversion from the forward mapping. The actual air braking force can be calculated from the actual brake cylinder pressure. Execution errors can be fed back to the car-to-car stability optimization layer to adjust weighting coefficients or correct prediction model parameters. Through these methods, a hierarchical closed-loop control can be formed, from the car-to-ground speed safety layer to the car-to-car stability optimization layer and then to the air-electric robust execution layer.

[0113] The feedforward compensation of the slave locomotive can be superimposed on the traction or braking force command of the slave locomotive to preemptively offset the resistance imbalance caused by gradient changes during the time delay. The future state of the slave locomotive in the asynchronous prediction model can be determined based on the slave locomotive's current command and the historical lagging command of the master locomotive. The asynchronous synchronization deviation index can quantify the degree of asynchronous timing between the master and slave control commands. The asynchronous rate constraint can force the change rate of the master and slave control commands to remain coordinated. The multi-objective cost function can include the square term of the coupler force at the slave locomotive, the penalty term for the rate of change of traction or electric braking force, the penalty term for the amount of air braking applied, and the penalty term for asynchronous synchronization deviation. The weight allocation principle can ensure that coupler force suppression is dominant, while taking into account both execution smoothness and asynchronous coordination. By solving through the model predictive control framework, the solution can be rolled over in each control cycle, synchronously outputting the optimal command sequence of traction or electric braking force and air braking force for the master locomotive and the slave locomotive respectively.

[0114] The robust air-electric braking layer leverages the fast response speed of the traction motor to design a rapid compensator based on air braking force execution error. The compensator can dynamically correct traction or electric braking force commands using a robust proportional-integral (PI) regulation law. The compensation mechanism can operate independently on the local controllers of both the master and slave locomotives. When air braking force is insufficient, electric braking force can be increased to compensate for the shortfall. When air braking force overshoots, electric braking force can be reduced to restore balance. The total actual output force of the master and slave locomotives can be the sum of their respective final electric braking force and actual air braking force. Through local closed-loop compensation, robust tracking of the actual output force to the vehicle-to-vehicle stability optimization layer's optimization commands can be ensured, and the steady-state tracking error can be controlled within allowable thresholds.

[0115] The actual total force can be output to the train actuators, while the execution error is fed back to the vehicle-to-vehicle stability optimization layer for parameter adjustment in subsequent optimization cycles. This forms a closed-loop hierarchical control. By hierarchically dividing the vehicle-to-ground speed safety layer and the vehicle-to-vehicle stability optimization layer, speed safety and stability optimization can be handled separately at different control levels, overcoming the mutual interference caused by multiple objectives coupled to the same decision layer in traditional architectures. By incorporating communication latency as a hard constraint into the optimization framework aimed at minimizing coupler force, and by including air braking force in the master-slave control differentiated optimization decision variables, the surge problem caused by asynchrony can be solved, enhancing the coordination consistency of master-slave control locomotives. By introducing asynchronous coordination constraints and asynchronous penalty terms, quantitative control of the timing differences between master-slave control commands can be achieved. By introducing a closed-loop robust compensation mechanism for air braking nonlinearity at the bottom layer, the lag of air braking can be compensated for, improving the overall control system's resistance to model errors, environmental changes, and execution aging.

[0116] See Figure 3 , Figure 3This application provides a schematic diagram of a multi-level cooperative control system for a heavy-haul combined train, which includes a data acquisition module, a vehicle-to-ground speed safety layer module, a vehicle-to-vehicle stability optimization layer module, and an air-electric robust execution layer module.

[0117] The data acquisition module can be used to acquire train status data, track data, real-time gradient of the main control locomotive and the real-time gradient of the slave control locomotive.

[0118] The vehicle-to-ground speed safety layer module can be used to calculate the total demand force of the train based on status data, track data, real-time gradient of the master locomotive and the slave locomotive, and prediction of the forward gradient based on the force balance of the train. It can also distribute the total demand force of the train differentially based on the gradient difference between the master locomotive and the slave locomotive, and output the distribution results. The distribution results include the safety boundary force of the master locomotive, the safety boundary force of the slave locomotive, the adjustment margin of the master locomotive, and the adjustment margin of the slave locomotive.

[0119] The vehicle-to-vehicle stability optimization layer module can be used to receive the allocation results, aim to minimize the coupler force at the slave locomotive, and construct a model predictive control framework containing a predictive model, with communication delay and power asynchrony characteristics as constraints. The model predictive control framework is used to optimize the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive, and output the optimal command sequence.

[0120] The air-electric robust execution layer module can be used to receive the optimal command sequence, determine the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determine the actual air braking force and execution error based on the actual brake cylinder pressure, correct the traction force or electric braking force command based on the execution error, output the braking command or final traction force, and feed the execution error back to the vehicle stability optimization layer; the braking command includes the final air braking command and / or electric braking force command; the execution error is used to correct the model parameters of the prediction model.

[0121] Based on the above embodiments, as a preferred embodiment, the vehicle-to-ground speed safety layer module may include a desired acceleration determination unit, a total resistance determination unit, a total demand force determination unit, a gradient difference determination unit, a safety boundary force determination unit, and a margin determination unit. The desired acceleration determination unit can be used to determine the desired acceleration based on the deviation between the current speed and the target speed. The total resistance determination unit can be used to determine the total resistance based on the total train mass, basic resistance coefficient, current speed, weighted average gradient, and curve radius. The total demand force determination unit can be used to determine the total demand force of the train based on the inertial force corresponding to the desired acceleration and the total resistance. The gradient difference determination unit can be used to determine the gradient difference based on the real-time gradient of the master locomotive and the slave locomotive. The safety boundary force determination unit can be used to determine the safety boundary force of the master locomotive and the safety boundary force of the slave locomotive based on the total demand force of the train, the locomotive quantity allocation coefficient, and the gradient difference. The margin determination unit can be used to determine the margin of the master locomotive and the margin of the slave locomotive based on a first set ratio of the maximum traction force or a second set ratio of the braking force.

[0122] Based on the above embodiments, as a preferred embodiment, the vehicle-to-vehicle stability optimization layer module may include a time-delay position prediction unit, an effective gradient difference determination unit, a feedforward compensation unit, an asynchronous prediction model establishment unit, an asynchronous synchronization deviation determination unit, an asynchronous rate constraint determination unit, a decision variable setting unit, a multi-objective cost function construction unit, and a solution output unit. The time-delay position prediction unit can be used to predict the position of the master locomotive and the position of the slave locomotive after the time delay based on communication delay, the current speed and acceleration of the master and slave locomotives. The effective gradient difference determination unit can be used to extract the gradient corresponding to the position of the master locomotive and the position of the slave locomotive after the time delay based on line data, and determine the effective gradient difference based on the two. The feedforward compensation unit can be used to determine the feedforward compensation amount of the slave locomotive based on the effective gradient difference, and superimpose the feedforward compensation amount of the slave locomotive onto the traction or braking force command of the slave locomotive. The asynchronous prediction model building unit can be used to establish an asynchronous prediction model based on the update cycle of the master locomotive's command and the slave locomotive's command. The future state of the slave locomotive in the asynchronous prediction model is determined based on the slave locomotive's current command and the master locomotive's historical lagging command. The asynchronous synchronization deviation determination unit can be used to determine the asynchronous synchronization deviation index based on the master locomotive's current command and the slave locomotive's historical command. The asynchronous rate constraint determination unit can be used to determine the asynchronous rate constraint based on the change rate of the master locomotive's current command and the change rate of the slave locomotive's historical command. The decision variable setting unit can be used to set the master locomotive's traction force or electric braking force, the slave locomotive's traction force or electric braking force, the master locomotive's air braking force, and the slave locomotive's air braking force as decision variables. The multi-objective cost function construction unit can be used to construct a multi-objective cost function, which includes a term for the square of the coupler force at the slave locomotive, a penalty term for the change rate of the master and slave locomotive's traction force or electric braking force, a penalty term for the amount of air braking applied by the master and slave locomotives, and a penalty term for the asynchronous synchronization deviation. The output unit can be used to solve for decision variables and obtain the optimal instruction sequence based on the train's longitudinal dynamic equations, the physical limits of the actuators, the safety threshold of the coupler force at the slave locomotive, the speed safety boundary constraints, the asynchronous coordination constraints, and the asynchronous rate constraints.

[0123] Based on the above embodiments, as a preferred embodiment, the vehicle-to-vehicle smoothing optimization layer module may further include a rolling solution unit, a smoothing deployment unit, and a recording feedback unit. The rolling solution unit can be used to incorporate multi-objective cost functions and constraints into the model predictive control framework, performing rolling solutions within each control cycle and outputting the optimal command sequences corresponding to the master locomotive and the slave locomotive, respectively. The smoothing deployment unit can be used to rate-limit and smoothly deploy the optimal command sequences according to the force change rate limit and the smoothing time constant; wherein the master locomotive and the slave locomotive have different smoothing time constants and deployment time offsets. The recording feedback unit can be used to record the asynchronous deviation and the coupler force response at the slave locomotive in the current cycle, for use in adjusting the parameters of the predictive model in the next optimization cycle.

[0124] Based on the above embodiments, as a preferred embodiment, the air-electric robust execution layer module may include an air brake model establishment unit, a nominal decompression determination unit, an actual air braking force determination unit, an execution error determination unit, a compensation amount determination unit, and a final command determination unit. The air brake model establishment unit can be used to establish a forward mapping between brake cylinder pressure and air braking force and a pre-trained air brake inverse model based on air brake system test data. The nominal decompression determination unit can be used to determine the nominal decompression based on the air braking force command in the optimal command sequence and the pre-trained air brake inverse model. The actual air braking force determination unit can be used to determine the actual air braking force based on the actual brake cylinder pressure measurement value and the forward mapping. The execution error determination unit can be used to determine the air braking force execution error based on the air braking force command and the actual air braking force. The compensation amount determination unit can be used to determine the traction force or electric braking force compensation amount using a proportional-integral control law based on the air braking force execution error. The final command determination unit can be used to superimpose the traction force or electric braking force compensation amount onto the traction force or electric braking force command in the optimal command sequence to obtain the final traction force or electric braking force command.

[0125] Based on the above embodiments, as a preferred embodiment, the air-electric robust execution layer module may further include a total force determination unit, a steady-state error determination unit, an output unit, and a feedback unit. The total force determination unit can be used to determine the actual total force of the master locomotive and the actual total force of the slave locomotive based on the final traction force or electric braking force command and the actual air braking force. The steady-state error determination unit can be used to determine the steady-state tracking error based on the sum of the traction force or electric braking force command and the air braking force command, the actual total force of the master locomotive, and the actual total force of the slave locomotive. The output unit can be used to output the actual total force of the master locomotive and the actual total force of the slave locomotive to the train actuator when the steady-state tracking error meets a preset threshold. The feedback unit can be used to feed back the air braking force execution error and the actual coupler force to the car-to-car stability optimization layer.

[0126] 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. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

[0127] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A multi-stage cooperative control method of a heavy-load combined train, characterized by, include: Acquire train status data, track data, real-time gradient of the main control locomotive, and real-time gradient of the slave control locomotive; The vehicle-to-ground speed safety layer calculates the total demand force of the train based on the status data, the track data, the real-time gradient of the master locomotive and the real-time gradient of the slave locomotive, and predicts the forward gradient based on the force balance of the train. It then allocates the total demand force of the train differentially based on the gradient difference between the real-time gradient of the master locomotive and the real-time gradient of the slave locomotive, and outputs the allocation result. The allocation results include the safety boundary force of the master locomotive, the safety boundary force of the slave locomotive, the adjustment margin of the master locomotive, and the adjustment margin of the slave locomotive; The vehicle-to-vehicle stability optimization layer receives the allocation results and constructs a model predictive control framework containing a predictive model, with the goal of minimizing the coupler force at the slave locomotive and the constraints of communication delay and power asynchrony characteristics. The model predictive control framework is used to optimize the traction force or electric braking force command and air braking force command of the master locomotive and slave locomotive, and outputs the optimal command sequence. The air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, corrects the traction force or electric braking force command based on the execution error, outputs the braking command or final traction force, and feeds the execution error back to the vehicle stability optimization layer; the braking command includes the final air braking command and / or electric braking force command; the execution error is used to correct the model parameters of the prediction model.

2. The method of claim 1, wherein, The vehicle-to-ground speed safety layer calculates the total train demand force based on the status data, the track data, the real-time gradient of the master locomotive, and the real-time gradient of the slave locomotive, using train force balance and forward gradient prediction. It then allocates the total train demand force differentially based on the gradient difference between the master and slave locomotives, outputting the allocation results including: The desired acceleration is determined based on the deviation between the current velocity and the target velocity; The total resistance is determined based on the total train mass, basic resistance coefficient, current speed, weighted average gradient, and curve radius. The total required force of the train is determined based on the inertial force corresponding to the desired acceleration and the total resistance. The gradient difference is determined based on the real-time gradient of the master locomotive and the real-time gradient of the slave locomotive; The safety boundary force of the master locomotive and the safety boundary force of the slave locomotive are determined based on the total train demand force, the locomotive quantity allocation coefficient and the gradient difference. The adjustment margin of the master locomotive and the adjustment margin of the slave locomotive are determined according to a first set ratio of the maximum traction force or a second set ratio of the braking force.

3. The method of claim 1, wherein, The model predictive control framework is used to optimize the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive, and the optimal command sequence is output, including: Based on the communication delay, the current speed and acceleration of the master locomotive and slave locomotive, predict the position of the master locomotive and the position of the slave locomotive after the delay; Extract the gradient corresponding to the position of the master locomotive after the time delay and the gradient corresponding to the position of the slave locomotive after the time delay from the line data, and determine the effective gradient difference based on the two. The feedforward compensation amount of the slave locomotive is determined based on the effective gradient difference, and the feedforward compensation amount of the slave locomotive is superimposed on the traction or braking force command of the slave locomotive. An asynchronous prediction model is established in the prediction model based on the master locomotive instruction update cycle and the slave locomotive instruction update cycle. The future state of the slave locomotive in the asynchronous prediction model is determined based on the slave locomotive's current instruction and the master locomotive's historical lagging instruction. The asynchronous synchronization deviation index is determined based on the current instructions of the master locomotive and the historical instructions of the slave locomotive. The asynchronous rate constraint is determined based on the current command change rate of the master locomotive and the historical command change rate of the slave locomotive.

4. The method of claim 3, wherein, The vehicle-to-vehicle stability optimization layer outputs the optimal instruction sequence, including: The traction force or electric braking force of the master locomotive, the traction force or electric braking force of the slave locomotive, the air braking force of the master locomotive, and the air braking force of the slave locomotive are used as decision variables. Construct a multi-objective cost function, which includes a term for the square of the coupler force at the slave locomotive, a penalty term for the rate of change of traction force or electric braking force of the master locomotive and the slave locomotive, a penalty term for the amount of air brake input of the master locomotive and the slave locomotive, and a penalty term for asynchronous synchronization deviation. Based on the train longitudinal dynamics equation, the physical limits of the actuator, the safety threshold of the coupler force at the slave locomotive, the speed safety boundary constraints determined by the safety boundary forces of the master locomotive, the safety boundary forces of the slave locomotive, the adjustment margins of the master locomotive and the slave locomotive, the asynchronous coordination constraints corresponding to the asynchronous-synchronous deviation index, and the asynchronous rate constraints, the decision variables are solved to obtain the optimal command sequence.

5. The method of claim 4, wherein, After outputting the optimal instruction sequence, the following is also included: The multi-objective cost function and constraints are incorporated into the model predictive control framework, and the solution is rolled over in each control cycle to output the optimal instruction sequence for the master locomotive and the slave locomotive respectively. The optimal command sequence is rate-limited and smoothly issued based on the force change rate limit and the smoothing time constant; wherein the master locomotive and the slave locomotive have different smoothing time constants and issuance time offsets. Record the asynchronous deviation and coupler force response at the slave locomotive in the current cycle so that the parameters of the prediction model can be adjusted for the next optimization cycle.

6. The method according to claim 1, characterized in that, The air-electric robust execution layer receives the optimal command sequence, determines the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determines the actual air braking force and execution error based on the actual brake cylinder pressure, and corrects the traction force or electric braking force command based on the execution error, including: Based on the test data of the air braking system, a positive mapping between brake cylinder pressure and air braking force and the pre-trained air braking inverse model are established. The nominal decompression is determined based on the air braking force command in the optimal command sequence and the pre-trained air braking inverse model. The actual air braking force is determined based on the actual brake cylinder pressure measurement and the positive mapping. The air braking force execution error is determined based on the air braking force command and the actual air braking force. The traction force or electric braking force compensation amount is determined based on the air braking force execution error. The traction force or electric braking force compensation is superimposed on the traction force or electric braking force command in the optimal command sequence to obtain the final traction force or electric braking force command.

7. The method according to claim 6, characterized in that, Outputting braking commands or final traction force, and feeding back the execution error to the vehicle stability optimization layer, including: The actual total force of the master locomotive and the actual total force of the slave locomotive are determined based on the final traction force or electric braking force command and the actual air braking force. The steady-state tracking error is determined based on the sum of the traction force or electric braking force command and the air braking force command, the actual total force of the master locomotive, and the actual total force of the slave locomotive. When the steady-state tracking error meets the preset threshold, the actual total force of the master locomotive and the actual total force of the slave locomotive are output to the train actuator. The air braking force execution error and the actual coupler force are fed back to the vehicle-to-vehicle stability optimization layer.

8. A multi-level cooperative control system for heavy-haul combined trains, characterized in that, include: The data acquisition module is used to acquire train status data, track data, real-time gradient of the main control locomotive, and real-time gradient of the slave control locomotive. The vehicle-to-ground speed safety layer module is used to calculate the total demand force of the train based on status data, track data, real-time gradient of the master locomotive and slave locomotive, and prediction of the forward gradient, based on the force balance of the train and the gradient ahead. It then distributes the total demand force of the train differentially according to the gradient difference between the master locomotive and slave locomotive, and outputs the distribution results. The distribution results include the safety boundary force of the master locomotive, the safety boundary force of the slave locomotive, the adjustment margin of the master locomotive, and the adjustment margin of the slave locomotive. The vehicle-to-vehicle stability optimization layer module receives the allocation results and, with the goal of minimizing the coupler force at the slave locomotive and constrained by communication delay and power asynchrony characteristics, constructs a model predictive control framework containing a predictive model. The model predictive control framework is used to optimize the traction or electric braking force commands and air braking force commands of the master locomotive and slave locomotive, and outputs the optimal command sequence. The air-electric robust execution layer module is used to receive the optimal command sequence, determine the nominal decompression amount based on the air braking force command and the pre-trained air braking inverse model, determine the actual air braking force and execution error based on the actual brake cylinder pressure, correct the traction force or electric braking force command based on the execution error, output the braking command or final traction force, and feed the execution error back to the vehicle stability optimization layer; the braking command includes the final air braking command and / or electric braking force command; the execution error is used to correct the model parameters of the prediction model.

9. The system according to claim 8, characterized in that, The vehicle-to-ground speed safety layer module includes: The desired acceleration determination unit can be used to determine the desired acceleration based on the deviation between the current velocity and the target velocity; The total resistance determination unit can be used to determine the total resistance based on the total train mass, basic resistance coefficient, current speed, weighted average gradient, and curve radius. The total demand force determination unit can be used to determine the total demand force of a train based on the inertial force and total resistance corresponding to the desired acceleration. The gradient difference determination unit can be used to determine the gradient difference based on the real-time gradient of the master locomotive and the real-time gradient of the slave locomotive; The safety boundary force determination unit can be used to determine the safety boundary force of the master locomotive and the safety boundary force of the slave locomotive based on the total demand force of the train, the locomotive quantity allocation coefficient and the gradient difference; The adjustment margin determination unit can be used to determine the adjustment margin of the master locomotive and the slave locomotive based on a first set ratio of the maximum traction force or a second set ratio of the braking force.

10. The system according to claim 8, characterized in that, The vehicle stability optimization layer module also includes: The rolling solver unit can be used to incorporate multi-objective cost functions and constraints into the model predictive control framework, and perform rolling solver in each control cycle to output the optimal instruction sequence for the master locomotive and the slave locomotive respectively. The smoothing delivery unit can be used to rate limit and smooth the delivery of the optimal command sequence based on the force change rate limit and the smoothing time constant; wherein the master locomotive and the slave locomotive have different smoothing time constants and delivery time offsets; The recording feedback unit can be used to record the asynchronous deviation of the current cycle and the coupler force response at the slave locomotive, so as to adjust the parameters of the prediction model for the next optimization cycle.