A distributed cooperative control method and system for a mass transit train and a medium

Abstract

The application discloses a kind of distributed cooperative control method, system and medium of collection and dispersion train, belong to the technical field of rail transit train operation control, this method includes: establishing the longitudinal dynamics model of collection and dispersion train;Based on car-car communication, the communication topology of collection and dispersion train is constructed, and the operation constraint condition of each train is determined;Based on communication topology and operation constraint condition, the local optimization problem of each train is constructed;Based on longitudinal dynamics model and local optimization problem, the local optimization problem is solved by using distributed model predictive control method rolling, the optimal control input sequence of each train in the prediction time domain is obtained, and the first control input of optimal control input sequence is applied to corresponding train.The application solves the technical problems that the existing collection and dispersion multi-train cooperative operation technology is difficult to balance operation safety and line logistics transportation efficiency due to large tracking interval, insufficient cooperative control precision, and difficulty in matching distributed autonomous decision and safety hard constraint.

Description

Technical Field

[0001] This invention belongs to the technical field of rail transit train operation control, specifically relating to a distributed collaborative control method, system, and medium for collection and distribution trains. Background Technology

[0002] With economic development, industrial restructuring, and accelerated consumption upgrading, in response to the "Action Plan for Effectively Reducing Logistics Costs Across Society" and related policies such as "Building a Strong Transportation Nation," and to adapt to the needs of modern collection and distribution logistics transportation development, collection and distribution transportation equipment needs to promote continuous technological innovation to significantly improve the modernization and intelligence level of collection and distribution freight transportation, promote the transformation and upgrading of my country's collection and distribution transportation towards intelligent transportation, and achieve an overall technical level that reaches the world's advanced level.

[0003] In traditional train control modes, trains operate independently, obtaining the position information of the preceding train only through the ground signaling system. This necessitates a large safety margin in train intervals, limiting track capacity and increasing the risk of rear-end collisions and delays due to unforeseen circumstances. Distributed cooperative control methods based on the train operation control system for shared and distributed rail transit, through real-time information exchange between trains and between trains and the ground, enable speed coordination, interval control, and operational optimization among multiple trains. This can shorten train intervals to tens of meters or even less while ensuring safety, significantly increasing track capacity. Simultaneously, cooperative control can reduce system energy consumption by approximately 10%-20% by optimizing train traction and braking strategies, resulting in significant economic and social benefits. Therefore, developing advanced train cooperative control technology is crucial for building a safe, efficient, and green modern rail transit system. Summary of the Invention

[0004] This invention aims to solve the core problems of existing multi-train cooperative operation technology for collection and distribution, which suffers from large tracking intervals, insufficient cooperative control precision, and difficulty in matching distributed autonomous decision-making with hard safety constraints, thus making it difficult to balance operational safety and line logistics transportation efficiency. The invention provides a distributed cooperative control method for collection and distribution trains.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a distributed cooperative control method for collecting and distributing trains, comprising: Establish a longitudinal dynamics model for the collection and distribution train; The communication topology of the collection and distribution trains is constructed based on vehicle-to-vehicle communication, and the operating constraints of each train are determined. Based on the communication topology and the operational constraints, a local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change in control input, and maximizing the consistency of the states of adjacent trains. Based on the longitudinal dynamics model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner to obtain the optimal control input sequence for each train in the prediction time domain. The first control input of the optimal control input sequence is then applied to the corresponding train to achieve coordinated control of the collection and distribution trains.

[0006] In one possible implementation, establishing the longitudinal dynamics model of the collection and distribution train includes: The longitudinal motion of each train is described using a single-mass model, and the longitudinal dynamic equations are established based on traction force, braking force, and running resistance. The running resistance is described using the Davis formula, and the Davis formula is linearized by a first-order Taylor expansion at the current average running speed. A traction / braking hysteresis coefficient is introduced to establish a first-order inertial link between the desired acceleration determined by the traction / braking command and the actual acceleration, so as to characterize the response delay characteristics of the traction system and the braking system. Using the position, velocity, and actual acceleration in the longitudinal dynamic equation as state variables and the traction / braking command as the control input, a continuous-time state-space model incorporating the response delay characteristics is established. The continuous-time state-space model is discretized to obtain a discrete-time longitudinal dynamic model.

[0007] In one possible implementation, the communication topology is a chain-type bidirectional communication topology, in which each train in the collection and distribution train only interacts with the adjacent train.

[0008] In one possible implementation, the operational constraints include state constraints, control input constraints, and coupling safety constraints; The state constraints include: at each sampling time in the prediction time domain, the train speed is within the speed feasible range determined by the line speed limit and the ATP speed limit, and the actual train acceleration is within the acceleration feasible range determined by the train's maximum acceleration and deceleration performance. The control input constraints include: at each sampling time in the prediction time domain, the traction / braking command as the control input is within the feasible range of the command amplitude determined by the maximum output capacity of the train traction / braking system, and the change in the traction / braking command between adjacent sampling times is within the feasible range of the rate of change determined by the system response characteristics. The coupled safety constraint includes: at each sampling time in the prediction time domain, the actual distance between adjacent trains is greater than or equal to the minimum safety distance.

[0009] In one possible implementation, the optimization objective includes a tracking error term, a control increment term, and an adjacent train consistency term; The tracking error term includes the sum of stage tracking cost and terminal tracking cost; the stage tracking cost is the sum of the weighted quadratic norms of the deviations between the train's predicted state and the reference running state at each prediction step in the prediction time domain; the terminal tracking cost is the weighted quadratic norm of the deviation between the train's predicted state and the terminal reference state at the end of the prediction time domain. The control increment term is determined based on the norm of the difference between control inputs at adjacent time points within the prediction time domain; The adjacent train consistency term is determined based on the predicted state difference between the current train and adjacent trains within the prediction time domain.

[0010] In one possible implementation, the local optimization problem includes: Using the state constraints, control input constraints, and coupled safety constraints as constraints, and minimizing the sum of the tracking error term, the control increment term, and the adjacent train consistency term as the optimization objective, a quadratic programming problem is solved in the prediction time domain to find the optimal control input sequence.

[0011] In one possible implementation, the method of using a distributed model predictive control to solve the local optimization problem in a rolling manner includes: Step 1: Set the initial system state at the initial moment, as well as the initial feasible control sequence and corresponding state sequence for each train; Step 2: At the current sampling time, each train generates a candidate input sequence and a corresponding candidate state sequence by superimposing a local auxiliary control law based on the optimal input sequence and the corresponding state sequence of the previous time. The local auxiliary control law is determined based on the predicted state difference between the train and the adjacent train and the corresponding preset feedback gain matrix. Step 3: Exchange the candidate state sequences of each train with those of its neighboring trains; Step 4: Based on the current state of each train and the candidate state sequence of its neighboring trains, solve the quadratic programming optimization problem of each train in parallel to obtain the optimal control input sequence and the corresponding optimal state sequence of each train in the prediction time domain. Step 5: Apply the first element of the optimal control input sequence as the control command at the current sampling time to the corresponding train; Step 6: Proceed to the next sampling time, return to step 2, and repeat steps 2 to 5 to achieve rolling optimization control.

[0012] Secondly, the present invention also provides a distributed cooperative control system for collecting and distributing trains, comprising: The ATS central control layer is used to issue reference operating status to each train and monitor the multi-train coordination status. The trackside layer is used to exchange information between the central motion control layer of the ATS and the on-board layer. In the onboard layer, a longitudinal dynamic model of the collection and distribution trains is established; a communication topology for the collection and distribution trains is constructed based on vehicle-to-vehicle communication, and the operating constraints of each train are determined; based on the communication topology and the operating constraints, a local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change in control input, and maximizing the consistency of the states of adjacent trains; based on the longitudinal dynamic model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner to obtain the optimal control input sequence for each train in the prediction time domain, and the first control input of the optimal control input sequence is applied to the corresponding train to achieve coordinated control of the collection and distribution trains.

[0013] In one possible implementation, the on-board layer includes an optimization controller, a state predictor, an automatic train operation unit, an automatic train protection and safety unit, and a low-level execution unit. The optimization controller is used to obtain the status of adjacent trains based on the communication topology constructed by vehicle-to-vehicle communication. It uses minimum spacing, speed limit and traction and braking performance as operating constraints. Based on the objective function and state deviation, it calculates the optimal control quantity and the optimal control input sequence based on the longitudinal dynamic model, and uses the optimal control input sequence as the input of the automatic train operation unit and the state predictor. The state predictor is used to receive the optimal control input sequence, combine it with the real-time train operation status fed back by the automatic train operation unit, predict the state trajectory in the prediction time domain based on the longitudinal dynamics model, perform differential calculation with the reference operation status, and transmit the obtained state deviation to the optimization controller. The automatic train operation unit is used to receive the optimal control input sequence and convert it into actual control commands, while simultaneously feeding back the real-time operating status of the train; The automatic train protection safety unit is used to perform safety verification on all control commands in advance and output hard constraints on traction and braking performance. The underlying execution unit is used to execute the actual control instructions and provide status feedback; The Automatic Train Operation Unit is an ATO unit with DMPC function, the Optimization Controller is a DMPC Optimization Controller, the State Predictor is a DMPC State Predictor, the Automatic Train Protection and Safety Unit is an ATP unit with AOM function, and the underlying execution unit includes CCU, TCU and BCU.

[0014] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the distributed cooperative control method for collection and distribution trains as described in any of the above method items.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The distributed predictive cooperative control method proposed in this invention can achieve multi-dimensional state consistency convergence of train position, speed, and acceleration. Under the premise of strictly meeting safety hard constraints, it can effectively shorten the train tracking interval, improve the efficiency of line logistics transportation, and at the same time ensure the smooth and stable control process, with good real-time performance and robustness. 2. This invention reduces the computational burden and communication dependence of the central node, and improves the fault redundancy capability and autonomous decision-making level of the multi-vehicle cooperative system. Attached Figure Description

[0016] Figure 1 This is a flowchart of a distributed cooperative control method for a collection and distribution train according to an embodiment of the present invention; Figure 2 This is an embodiment of the present invention. Figure 1 Flowchart of step S1; Figure 3 This is an embodiment of the present invention. Figure 1 Flowchart of step S4; Figure 4 This is a structural diagram of the distributed collaborative control system for the collection and distribution trains according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the distributed predictive control performed by the distributed cooperative control system according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] The present invention will be further described below with reference to specific embodiments, but these are not intended to limit the scope of the invention.

[0020] Example 1 Firstly, such as Figure 1 This invention provides a distributed cooperative control method for collecting and distributing trains, comprising: S1. Establish a longitudinal dynamics model for the collection and distribution trains; S2. Construct the communication topology of the collection and distribution trains based on vehicle-to-vehicle communication, and determine the operating constraints of each train; S3. Based on the communication topology and operational constraints, the local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change of control input, and maximizing the consistency of the states of adjacent trains. S4. Based on the longitudinal dynamics model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner, so as to obtain the optimal control input sequence of each train in the prediction time domain, and the first control input of the optimal control input sequence is applied to the corresponding train to achieve coordinated control of the collection and distribution trains.

[0021] To address the technical problems of existing multi-train cooperative operation technologies for passenger transport, such as large tracking intervals, insufficient cooperative control precision, and difficulty in matching distributed autonomous decision-making with hard safety constraints, which makes it difficult to balance operational safety and line logistics efficiency, this invention establishes a longitudinal dynamic model for passenger transport trains. Using a single-mass model combined with the Davis resistance formula and traction / braking response delay, a predictive model is constructed for each train, combining physical accuracy with computational simplicity. This allows each train to predict future multi-step motion trends based on its current state, laying the foundation for high-precision cooperative control. Furthermore, a chain-like bidirectional communication topology is constructed based on vehicle-to-vehicle communication, where each train only interacts with adjacent trains. The communication volume is linearly related to the number of trains, significantly reducing communication bandwidth pressure while meeting the information requirements of cooperative control. Operational constraints, including state constraints, control input constraints, and coupled safety constraints, are defined. Safety requirements such as speed limits, acceleration limits, command amplitude limits, and minimum safe distance between adjacent trains are explicitly expressed in the form of inequality hard constraints. Finally, the minimum tracking error, minimum control increment, and minimum adjacent train state are considered as the criteria for achieving high-precision cooperative control. The optimal convergence principle is a multi-objective joint optimization objective. With operational constraints as the basis, a local optimization problem is constructed for each train, involving only information about the train itself and adjacent trains. This allows each train to solve the problem independently and in parallel, achieving distributed autonomous decision-making. Simultaneously, safety constraints, as hard constraints of inequalities in the optimization problem, are embedded in the solution process. Any feasible solution output by the solver satisfies safety requirements, fundamentally resolving the contradiction between distributed autonomous decision-making and hard safety constraints. Finally, a distributed model predictive control method is used to solve the local optimization problem of each train at each sampling time, re-optimizing based on the latest measured state. Online self-correction is performed for model mismatch and external disturbances. The first control input of the optimal control input sequence is applied to the corresponding train. Through multi-objective collaborative optimization, each train operates stably while maintaining the minimum safe distance as close as possible, compressing the actual tracking interval to near the theoretical minimum, effectively suppressing speed oscillations and spacing fluctuations. From the three dimensions of shortened tracking interval, improved collaborative control accuracy, and hard safety constraint guarantee under distributed decision-making, both operational safety and line logistics transportation efficiency are considered.

[0022] Regarding step S1, in one possible implementation, such as Figure 2 Establish a longitudinal dynamics model for the collection and distribution trains, including: S11. The longitudinal motion of each train is described using a single-mass model, and the longitudinal dynamic equations are established based on traction force, braking force, and running resistance.

[0023] Specifically, a single-mass model is used to describe the longitudinal motion of each train, treating the entire train as a concentrated mass body, and establishing the longitudinal dynamic equations based on Newton's second law: (1) In equation (1), For train quality, For the train in t The actual acceleration at time t. for t The traction or braking force at any given moment. For the train in t The basic operating resistance at any given time, for t Additional resistance at any given moment.

[0024] S12. The Davis formula is used to describe the running resistance, and the Davis formula is linearized by a first-order Taylor expansion at the current average running speed.

[0025] Specifically, the basic operating resistance of the train is calculated using the Davis formula, which is currently the most widely used method for resistance calculation in engineering: (2) In equation (2), The Davis drag coefficient reflects the effects of wheel-rail friction, impact vibration, and air resistance, respectively. For the train in t The speed of time; This is the acceleration due to gravity.

[0026] Since Davis resistance is nonlinear, to facilitate controller design, at the current average operating speed... The first-order Taylor expansion of Davis's formula is as follows: (3) in, .

[0027] By approximating the nonlinear resistance as a linear function of velocity, the overall optimization problem can be reduced to a quadratic programming problem, significantly reducing the computational complexity of online calculations.

[0028] S13. Introduce a traction / braking hysteresis coefficient to establish a first-order inertial element between the desired acceleration determined by the traction / braking command and the actual acceleration, in order to characterize the response delay characteristics of the traction system and the braking system.

[0029] Specifically, actual traction and braking systems exhibit response delays; the actual acceleration cannot instantaneously reach the desired value after a control command is issued. To characterize this delay, a traction / braking hysteresis coefficient is introduced. Describe the dynamic process from desired acceleration to actual acceleration: (4) In equation (4), The desired acceleration command output by the controller.

[0030] This first-order inertial element enables the model to reflect the dynamic response characteristics of the actuator, avoiding excessive control commands or insufficient safety clearance caused by the controller ignoring delays.

[0031] S14. Using the position, velocity, and actual acceleration in the longitudinal dynamic equation as state variables and the traction / braking command as the control input, establish a continuous-time state-space model that includes response delay characteristics.

[0032] Preferably, the location of the train is selected. ,speed and actual acceleration As a state variable ,Right now The control input is the desired acceleration. The continuous-time state-space model of the train is then: (5) S15. Discretize the continuous-time state-space model to obtain the discrete-time longitudinal dynamic model.

[0033] Specifically, the zero-order preserve method is used to discretize the above continuous model, and the sampling time is... Thus, the discrete-time state-space model is obtained: (6) in, , (7) For a train group system consisting of N trains, its overall state space model is the direct sum of the individual train models: (8) In equation (8), Let be the system's total state vector. This is the overall control input vector for the system.

[0034] Regarding step S2, in one possible implementation, the communication topology is a chain-type bidirectional communication topology, where each train in the collection and distribution train only interacts with its adjacent train.

[0035] In the train system for distribution and collection, trains are arranged in operating sequence. Each train communicates bidirectionally with its preceding and following trains via vehicle-to-vehicle (V2V) wireless communication. Specifically, train i can receive status information (including position, speed, acceleration, and predicted state sequence) from trains i-1 and i+1, while simultaneously sending its own status information to both trains i-1 and i+1. This chain-like bidirectional topology has N-1 communication links (where N is the total number of trains), and the communication volume is linearly related to the number of trains. Each train only needs to maintain two communication links, simplifying communication management. Information propagates bidirectionally along the links, allowing any train to obtain the complete status of its preceding and following trains, meeting the information requirements for coordinated control. Adding or removing trains only requires inserting or deleting a node in the chain-like topology, without needing to reconstruct the global communication structure, adapting to the dynamic changes in train formation required in distribution and collection scenarios.

[0036] Specifically, a communication topology for the train group is constructed based on graph theory, treating each train as a vertex in the graph and the communication links between trains as edges. In the urban rail train tracking operation scenario, trains travel in one direction along the line, and each train only needs to communicate with the train in front of it; therefore, a chain-like bidirectional communication topology is adopted.

[0037] For a train group consisting of 3 trains, the correlation matrix of its communication topology and Laplace matrix They are respectively: (9) Define the state difference variable between the two vehicles as: (10) In equation (10), For Kronecker product, Let n be an n-order identity matrix, where n=3 represents the state dimension of a single train. Then... This indicates the state difference between train 1 and train 2. This represents the state difference between train 2 and train 3. The goal of coordinated control is to minimize this state difference. Converging to zero achieves state consistency for the train group.

[0038] In one possible implementation, the operational constraints include state constraints, control input constraints, and coupling safety constraints; State constraints include: at each sampling time in the prediction time domain, the train speed is within the speed feasible range determined by the line speed limit and the ATP speed limit, and the actual train acceleration is within the acceleration feasible range determined by the train's maximum acceleration and deceleration performance; The control input constraints include: at each sampling time in the prediction time domain, the traction / braking command as the control input is within the feasible range of the command amplitude determined by the maximum output capacity of the train traction / braking system, and the change in the traction / braking command between adjacent sampling times is within the feasible range of the rate of change determined by the system response characteristics; The coupled safety constraints include: at each sampling time in the prediction time domain, the actual distance between adjacent trains is greater than or equal to the minimum safety distance.

[0039] Specifically, to ensure the safe and stable operation of the train, the system must meet the following three types of constraints: (1) State constraints: The train's operating speed cannot exceed the track speed limit, its acceleration cannot exceed its traction / braking capacity, and its position cannot exceed the track end point. That is: (11) (12) (13) In equations (11)-(13), For train i In position Speed ​​limits are in place at that location. and These are the train's maximum traction acceleration and maximum braking deceleration, respectively. This represents the total length of the line.

[0040] (2) Control input constraint: The desired acceleration command output by the controller cannot exceed the output capacity of the train's traction / braking system. That is: (14) in, .

[0041] (3) Coupled safety constraints: To prevent rear-end collisions, a minimum safe distance must be maintained between the front of the following train and the rear of the preceding train. The positional error between adjacent trains is defined as: (15) In equation (15), Let i be the position of train i. Let i-1 be the position of train i adjacent to train i. For the length of the train, This is the minimum safety margin.

[0042] Coupling security constraints: (16) when At that time, the actual distance between the front of the following vehicle and the rear of the preceding vehicle was exactly 1 / 3. When a critical safety state is reached; At that time, the two vehicles collided from behind.

[0043] For step S3, in one possible implementation, the optimization objectives include a tracking error term, a control increment term, and an adjacent train consistency term; The tracking error term includes the sum of stage tracking cost and terminal tracking cost; stage tracking cost is the sum of the weighted quadratic norms of the deviations between the train's predicted state and the reference running state at each prediction step in the prediction time domain; terminal tracking cost is the weighted quadratic norm of the deviation between the train's predicted state and the terminal reference state at the end of the prediction time domain. The control increment term is determined based on the norm of the difference between control inputs at adjacent time points within the prediction time domain; The adjacent train consistency term is determined based on the difference in predicted state between the current train and adjacent trains within the prediction time domain.

[0044] Specifically, the overall objective function constructed in this embodiment includes three parts: tracking error, control increment, and adjacent train consistency cost.

[0045] For the i-th train, the optimization problem at time k is: (17) In equation (17), , which is the control input sequence in the prediction time domain. For prediction in the time domain.

[0046] (1) Tracking error term: used to measure the accuracy of the train tracking the global target speed curve.

[0047] (18) In equation (18), Let i be the reference state of train i at the predicted time k+l. Q and P These are the preset state weight matrix and terminal weight matrix, respectively.

[0048] (2) Control increment: used to punish drastic changes in control quantity, ensure smooth train operation, and improve stability.

[0049] (19) in, For train i at the predicted time k+ l control quantity ，R To control the input weight matrix.

[0050] (3) Consistency between adjacent trains: This is used to measure the consistency between the state of a train and its neighboring trains, and to ensure that trains maintain a safe and stable interval.

[0051] It should be noted that this is the core difference between DMPC cooperative control and traditional follower control: (20) in, For collaborative weights, For predicting the time k+l, train i and the preceding train i 1 is in a poor state.

[0052] In one possible implementation, the local optimization problem includes: Using state constraints, control input constraints, and coupled safety constraints as constraints, and minimizing the sum of the tracking error term, control increment term, and adjacent train consistency term as the optimization objective, a quadratic programming problem is solved in the prediction time domain to find the optimal control input sequence.

[0053] Regarding the specific form of the local optimization problem in step S3, specifically, the local optimization problem for each train is reduced to a standard quadratic programming (QP) problem: using the control input sequence as the decision variable, and... The goal is to minimize the input, with the longitudinal dynamic equation as the equality constraint and position, velocity, acceleration, command amplitude, and coupling safety constraints as the inequality constraints. Since the objective function is a quadratic function of the control input sequence and the constraints are linear inequalities, this problem is a standard convex quadratic programming problem. Mature algorithms such as the interior point method, effective set method, or ADMM can be used to solve it efficiently, ensuring the existence and uniqueness of the global optimal solution.

[0054] Regarding step S4, in one possible implementation, such as Figure 3 A distributed model predictive control method is used to solve local optimization problems in a rolling manner, including: S41. Obtain the reference operating state, set the initial system state at the initial moment, and the initial feasible control sequence and corresponding state sequence for each train.

[0055] Preferably, the reference operating state in this embodiment is represented in the form of a speed curve.

[0056] S42. At the current sampling time, each train generates a candidate input sequence and a corresponding candidate state sequence at the current time by superimposing a local auxiliary control law based on the optimal input sequence and the corresponding state sequence of the previous time. The local auxiliary control law is determined based on the predicted state difference between the train and the adjacent train and the corresponding preset feedback gain matrix. S43. Exchange the candidate state sequences of each train with those of its neighboring trains; S44. Based on the current state of each train and the candidate state sequence of its neighboring trains, solve the quadratic programming optimization problem of each train in parallel to obtain the optimal control input sequence and the corresponding optimal state sequence of each train in the prediction time domain. S45. Apply the first element of the optimal control input sequence as the control command at the current sampling time to the corresponding train; S46. Proceed to the next sampling time, return to step S42, and repeat S42-S45 to achieve rolling optimization control.

[0057] Specifically, the Distributed Model Predictive Control (DMPC) algorithm employs a rolling time-domain optimization strategy. At each sampling time, each train solves its own local optimization problem in parallel and exchanges predicted state information through neighborhood communication to iteratively update control commands. The specific solution steps are as follows: (1) Step 1: Initialization Initialize parameters such as speed limit, gradient, and curve radius to generate a global target speed curve. Set the initial time. k =0, initial state of the system Initial feasible control sequence for each train and the corresponding state sequence The initial state sequence of each train is sent to the neighboring train.

[0058] (2) Step 2: Generate candidate input sequences At any moment k Take the optimal input sequence from the previous time step. and state sequence Superimposed local auxiliary control law Generate the candidate input sequence for the current time step. and the corresponding candidate state sequence .

[0059] The local auxiliary control law is designed as follows: (twenty one) In equation (21), Let i be the set of neighbors of train i. This is the feedback gain matrix.

[0060] (3) Step 3: Exchange prediction information The generated candidate state sequences of each train Send to all neighboring vehicles and receive candidate state sequences from neighboring vehicles. , .

[0061] (4) Step 4: Parallel solution of the optimization problem Based on the individual states of each train and the predicted states of its neighboring trains, the quadratic programming (QP) optimization problems of each train are solved in parallel to obtain the optimal control input sequence. and the corresponding optimal state sequence .

[0062] (5) Step 5: Update control commands Each train will use the first element of the optimal control input sequence The current control command is output to the traction / braking system.

[0063] (6) Step 6: Rolling Iteration Let k = k + 1, return to step 2, and proceed to the optimization calculation for the next sampling period.

[0064] Secondly, such as Figure 4 The present invention also provides a distributed cooperative control system 40 for collecting and distributing trains, comprising: The ATS central control layer 410 is used to issue reference operating status to each train and monitor the multi-train coordination status. The trackside layer 420 is used for information exchange between the ATS central control layer 410 and the on-board layer 430. In the onboard layer 430, a longitudinal dynamic model of the collection and distribution train is established; a communication topology of the collection and distribution train is constructed based on vehicle-to-vehicle communication, and the operating constraints of each train are determined; based on the communication topology and the operating constraints, a local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change of control input, and maximizing the consistency of the states of adjacent trains; based on the longitudinal dynamic model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner to obtain the optimal control input sequence of each train in the prediction time domain, and the first control input of the optimal control input sequence is applied to the corresponding train to achieve coordinated control of the collection and distribution train.

[0065] In one possible implementation, the on-board layer 430 includes an optimization controller, a state predictor, an automatic train operation unit, an automatic train protection and safety unit, and a low-level execution unit. The optimization controller is used to obtain the status of adjacent trains based on the communication topology constructed by vehicle-to-vehicle communication. It uses minimum spacing, speed limit and traction and braking performance as operating constraints. Based on the objective function and state deviation, it calculates the optimal control quantity and the optimal control input sequence based on the longitudinal dynamic model, and uses the optimal control input sequence as the input of the automatic train operation unit and the state predictor. The state predictor is used to receive the optimal control input sequence, combine it with the real-time train operation status fed back by the automatic train operation unit, predict the state trajectory in the prediction time domain based on the longitudinal dynamics model, perform differential calculation with the reference operation status, and transmit the obtained state deviation to the optimization controller. The automatic train operation unit is used to receive the optimal control input sequence and convert it into actual control commands, while also feeding back the real-time operating status of the train; The automatic train protection safety unit is used to perform safety verification on all control commands and output hard constraints on traction and braking performance. The underlying execution unit is used to execute the actual control instructions and provide status feedback; The Automatic Train Operation Unit (ATO) is an ATO unit with DMPC functionality, the Optimization Controller is a DMPC Optimization Controller, the State Predictor is a DMPC State Predictor, the Automatic Train Protection and Safety Unit (ATP) is an ATP unit with AOM functionality, and the underlying execution units include CCU, TCU, and BCU.

[0066] Specifically, such as Figure 5 The diagram illustrates the integrated structure of distributed predictive control and a simplified train operation control system. It employs a hierarchical design of "safety monitoring - optimization decision-making - state prediction." All control commands must first pass ATP verification, outputting hard constraints on traction and braking performance to prevent speeding, rear-end collisions, and other safety accidents at their source. ATO receives the optimized control sequence and transforms it into actual control commands, while simultaneously feeding back the train's real-time operating status. The optimization controller fully considers constraints such as minimum spacing, speed limits, and traction and braking performance, calculating the optimal control quantity and optimal control sequence based on the objective function and state deviation, serving as inputs to the ATO and the state predictor. The state predictor receives the optimal control sequence, combines it with the ATO to predict the state trajectory in the prediction time domain, performs differential calculations with the desired state, and transmits the state deviation to the optimization controller. The vehicle's underlying execution unit includes three core units: CCU, TCU, and BCU, responsible for executing control commands and providing state feedback.

[0067] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the distributed cooperative control method for collection and distribution trains as described in any of the above method items.

[0068] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the content of this specification should be included within the protection scope of the present invention.

Claims

1. A distributed cooperative control method for collection and distribution trains, characterized in that, include: Establish a longitudinal dynamics model for the collection and distribution train; The communication topology of the collection and distribution trains is constructed based on vehicle-to-vehicle communication, and the operating constraints of each train are determined. Based on the communication topology and the operational constraints, a local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change in control input, and maximizing the consistency of the states of adjacent trains. Based on the longitudinal dynamics model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner to obtain the optimal control input sequence for each train in the prediction time domain. The first control input of the optimal control input sequence is then applied to the corresponding train to achieve coordinated control of the collection and distribution trains.

2. The distributed cooperative control method for collection and distribution trains according to claim 1, characterized in that, The establishment of the longitudinal dynamics model for the collection and distribution trains includes: The longitudinal motion of each train is described using a single-mass model, and the longitudinal dynamic equations are established based on traction force, braking force, and running resistance. The running resistance is described using the Davis formula, and the Davis formula is linearized by a first-order Taylor expansion at the current average running speed. A traction / braking hysteresis coefficient is introduced to establish a first-order inertial link between the desired acceleration determined by the traction / braking command and the actual acceleration, so as to characterize the response delay characteristics of the traction system and the braking system. Using the position, velocity, and actual acceleration in the longitudinal dynamic equation as state variables and the traction / braking command as the control input, a continuous-time state-space model incorporating the response delay characteristics is established. The continuous-time state-space model is discretized to obtain a discrete-time longitudinal dynamic model.

3. The distributed cooperative control method for collection and distribution trains according to claim 1, characterized in that, The communication topology is a chain-type bidirectional communication topology, in which each train in the collection and distribution train only interacts with the adjacent train.

4. The distributed cooperative control method for collection and distribution trains according to claim 1, characterized in that, The operational constraints include state constraints, control input constraints, and coupling security constraints; The state constraints include: at each sampling time in the prediction time domain, the train speed is within the speed feasible range determined by the line speed limit and the ATP speed limit, and the actual train acceleration is within the acceleration feasible range determined by the train's maximum acceleration and deceleration performance. The control input constraints include: at each sampling time in the prediction time domain, the traction / braking command as the control input is within the feasible range of the command amplitude determined by the maximum output capacity of the train traction / braking system, and the change in the traction / braking command between adjacent sampling times is within the feasible range of the rate of change determined by the system response characteristics. The coupled safety constraint includes: at each sampling time in the prediction time domain, the actual distance between adjacent trains is greater than or equal to the minimum safety distance.

5. The distributed cooperative control method for collection and distribution trains according to claim 4, characterized in that, The optimization objectives include tracking error term, control increment term, and adjacent train consistency term; The tracking error term includes the sum of stage tracking cost and terminal tracking cost; The stage tracking cost is the sum of the weighted quadratic norms of the deviations between the train's predicted state and the reference operating state at each prediction step within the prediction time domain; the terminal tracking cost is the weighted quadratic norm of the deviation between the train's predicted state and the terminal reference state at the end of the prediction time domain. The control increment term is determined based on the norm of the difference between control inputs at adjacent time points within the prediction time domain; The adjacent train consistency term is determined based on the predicted state difference between the current train and adjacent trains within the prediction time domain.

6. The distributed cooperative control method for collection and distribution trains according to claim 5, characterized in that, The local optimization problem includes: Using the state constraints, control input constraints, and coupled safety constraints as constraints, and minimizing the sum of the tracking error term, the control increment term, and the adjacent train consistency term as the optimization objective, a quadratic programming problem is solved in the prediction time domain to find the optimal control input sequence.

7. The distributed cooperative control method for collection and distribution trains according to claim 6, characterized in that, The method of using distributed model predictive control to solve the local optimization problem in a rolling manner includes: Step 1: Set the initial system state at the initial moment, as well as the initial feasible control sequence and corresponding state sequence for each train; Step 2: At the current sampling time, each train generates a candidate input sequence and a corresponding candidate state sequence by superimposing a local auxiliary control law based on the optimal input sequence and the corresponding state sequence of the previous time. The local auxiliary control law is determined based on the predicted state difference between the train and the adjacent train and the corresponding preset feedback gain matrix. Step 3: Exchange the candidate state sequences of each train with those of its neighboring trains; Step 4: Based on the current state of each train and the candidate state sequence of its neighboring trains, solve the quadratic programming optimization problem of each train in parallel to obtain the optimal control input sequence and the corresponding optimal state sequence of each train in the prediction time domain. Step 5: Apply the first element of the optimal control input sequence as the control command at the current sampling time to the corresponding train; Step 6: Proceed to the next sampling time, return to step 2, and repeat steps 2 to 5 to achieve rolling optimization control.

8. A distributed cooperative control system for collecting and distributing trains, characterized in that, include: The ATS central control layer is used to issue reference operating status to each train and monitor the multi-train coordination status. The trackside layer is used to exchange information between the central motion control layer of the ATS and the on-board layer. In the onboard layer, a longitudinal dynamic model of the collection and distribution trains is established; a communication topology for the collection and distribution trains is constructed based on vehicle-to-vehicle communication, and the operating constraints of each train are determined; based on the communication topology and the operating constraints, a local optimization problem for each train is constructed with the optimization objectives of minimizing the error of each train tracking the reference operating state, minimizing the change in control input, and maximizing the consistency of the states of adjacent trains; based on the longitudinal dynamic model and the local optimization problem, a distributed model predictive control method is used to solve the local optimization problem in a rolling manner to obtain the optimal control input sequence for each train in the prediction time domain, and the first control input of the optimal control input sequence is applied to the corresponding train to achieve coordinated control of the collection and distribution trains.

9. The distributed cooperative control system for collection and distribution trains according to claim 8, characterized in that, The on-board layer includes an optimization controller, a state predictor, an automatic train operation unit, an automatic train protection and safety unit, and a bottom-level execution unit; The optimization controller is used to obtain the status of adjacent trains based on the communication topology constructed by vehicle-to-vehicle communication. It uses minimum spacing, speed limit and traction and braking performance as operating constraints. Based on the objective function and state deviation, it calculates the optimal control quantity and the optimal control input sequence based on the longitudinal dynamic model, and uses the optimal control input sequence as the input of the automatic train operation unit and the state predictor. The state predictor is used to receive the optimal control input sequence, combine it with the real-time train operation status fed back by the automatic train operation unit, predict the state trajectory in the prediction time domain based on the longitudinal dynamics model, perform differential calculation with the reference operation status, and transmit the obtained state deviation to the optimization controller. The automatic train operation unit is used to receive the optimal control input sequence and convert it into actual control commands, while simultaneously feeding back the real-time operating status of the train; The automatic train protection safety unit is used to perform safety verification on all control commands in advance and output hard constraints on traction and braking performance. The underlying execution unit is used to execute the actual control instructions and provide status feedback; The Automatic Train Operation Unit is an ATO unit with DMPC function, the Optimization Controller is a DMPC Optimization Controller, the State Predictor is a DMPC State Predictor, the Automatic Train Protection and Safety Unit is an ATP unit with AOM function, and the underlying execution unit includes CCU, TCU and BCU.

10. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, implements the distributed collaborative control method for the collection and distribution trains as described in any one of claims 1 to 7.

Images