Multi-locomotive traction heavy-duty train long and large downhill optimization method and system and storage medium

Abstract

The invention discloses a long and large downhill optimization method and system for a multi-locomotive traction heavy-load train and a storage medium, and relates to the technical field of automatic operation control of heavy-load trains. The method comprises the following steps: acquiring train parameter information, train characteristic parameters and line information, constructing a heavy-load train longitudinal dynamics multi-mass-point model according to the train parameter information, establishing a neural network and a heavy-load train long and large downhill multi-objective optimization reward function based on the train characteristic parameters and the heavy-load train longitudinal dynamics multi-mass-point model, and calculating the time when the train runs to the end. And comparing an output value of the heavy-load train long and large downhill multi-objective optimization reward function with a reward value generated by the neural network to construct a loss function, iteratively updating neural network parameters by using the loss function, calculating an Euclidean distance of the neural network parameters before and after updating, and stopping updating when a preset condition is reached to obtain an optimization control strategy. According to the invention, safe, efficient and stable operation of the heavy haul train can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic operation control of heavy-load trains, and in particular to a method, system and storage medium for optimizing a long downhill slope of a multi-locomotive traction heavy-load train. Background Art

[0002] Unlike passenger trains, heavy-load trains are mostly located in mountainous areas and are often accompanied by long and steep slopes. The line conditions are very bad. Taking the Shuohuang Railway as an example, the longitudinal section of the line is generally high in the west and low in the east, with an altitude difference of nearly 2,000 meters. The maximum slope in the direction of heavy vehicles is -12‰. When the train passes through a long downhill slope, electric braking alone cannot control the train deceleration. It is necessary to use a cyclic braking method and apply air braking for effective speed control. Due to the long formation and heavy weight of heavy-load trains, the stress conditions of each vehicle during operation are complex according to the line conditions. Unreasonable operation can easily lead to safety risks such as broken hooks and unhooking, which seriously affects transportation efficiency and causes huge economic losses and social impacts.

[0003] In order to ensure the transportation capacity of bulk goods, meet the growing demand for load weight, and connect various types of freight trains to improve transportation efficiency, heavy-duty trains use a multi-locomotive distributed traction method to improve traction quality. Due to the long formation of heavy-duty trains, when in braking conditions, the propagation time of the brake wave generated by the centralized traction method is longer than that of the distributed traction method, resulting in inconsistent operating states of each vehicle and a large longitudinal impact force between the vehicles. In particular, the central locomotive is often affected by a large longitudinal force. In severe cases, there may even be accidents such as vehicle derailment and coupler breakage, affecting the safety of train operation. To solve the above problems, it is necessary to study the change law of the characteristics of each carriage of the heavy-duty train when multiple locomotives are used for traction, and to study a universal optimization method for heavy-duty trains on long and steep downhill slopes.

[0004] Therefore, it is an urgent problem for technical personnel in this field to propose an optimization method, system and storage medium for multi-locomotive traction of heavy-load trains on long downhill slopes to solve the problems existing in the prior art. Summary of the invention

[0005] In view of this, the present invention provides a method, system and storage medium for optimizing a long downhill slope for a multi-locomotive traction heavy-load train, which helps to achieve safe, efficient and smooth operation of the heavy-load train.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] The optimization method for a multi-locomotive traction heavy-load train going downhill comprises the following steps:

[0008] S1. Obtain train parameter information, train characteristic parameters and line information;

[0009] S2. Construct a multi-particle model of the longitudinal dynamics of a heavy-load train based on train parameter information;

[0010] S3. Establish a neural network and a multi-objective optimization reward function for a long downhill heavy-haul train based on train characteristic parameters and a multi-mass point model of the longitudinal dynamics of a heavy-haul train;

[0011] S4, calculating the train operation until termination, comparing the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and using the loss function to iteratively update the neural network parameters;

[0012] S5. Calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are met, and obtain the optimized control strategy.

[0013] The above method optionally comprises constructing a multi-particle model of the longitudinal dynamics of a heavy-loaded train in S2, including: respectively constructing a basic resistance model based on traction, electric braking force, air resistance and additional resistance, a coupler force model between locomotive and truck, truck and truck, and an air brake time-lag model; and coupling the basic resistance model, the coupler force model and the air brake time-lag model to generate a multi-particle model of the longitudinal dynamics of a heavy-loaded train.

[0014] In the above method, the optional expression of the multi-mass point model of the longitudinal dynamics of heavy-duty train is:

[0015]

[0016] Among them, m i is the mass of the i-th carriage; v is the train speed; μ t,i is the traction output rate, μ t,i ∈[0,1], non-locomotive carriages take 0; F t,i is the train traction force; μ eb,i is the electric brake output rate, μ eb,i ∈[0,1], non-locomotive carriages take 0; F eb,i is the electric braking force of the train; μ fb,i Whether to output air brake, the value is 0 or 1; F fb,i F is the electric braking force of the train; c,i-1 is the coupler force generated by the (i-1) carriage; F c,i is the coupler force generated by the (i+1)th carriage; W 1,i W is the basic resistance encountered by the train during operation; 2,i It is the additional resistance encountered by the train during operation.

[0017] In the above method, optionally, the neural network in S3 includes an input layer, an intermediate layer and an output layer, and the feature vector formed by normalizing the train feature parameters in the input layer is used as the input vector of the neural network; the intermediate layer outputs the reward value corresponding to the train traction / braking.

[0018] In the above method, optionally, the multi-objective optimization reward function for the long downhill slope of the heavy-load train in S3 takes punctuality, safety, and brake device loss as optimization objectives.

[0019] In the above method, optionally, calculating the train operation in S4 includes:

[0020] S41, the neural network inputs the feature vector corresponding to the train entry point;

[0021] S42, obtaining a reward value corresponding to train traction / braking through calculation in the middle layer of the neural network;

[0022] S43, using the heavy-load train longitudinal dynamics multi-particle model and line information to calculate the running state of the train at the next node, determine whether it exceeds the speed limit or reaches the off-slope point, if not, construct a feature vector with the current node train state information, and repeat steps S42-S43;

[0023] S44. If yes, stop the calculation and distribute the corresponding reward.

[0024] In the above method, optionally, the preset condition in S5 includes: the Euclidean distance is less than a certain threshold or the maximum number of iterations is reached.

[0025] A multi-locomotive traction heavy-load train long downhill optimization system, which executes any of the multi-locomotive traction heavy-load train long downhill optimization methods described above, comprising an initialization module, a dynamic model construction module, a neural network construction module, an iterative optimization module and a strategy generation module connected in sequence;

[0026] Initialization module, used to input train parameter information, train characteristic parameters and line information;

[0027] Dynamic model building module, used to build a multi-particle model of the longitudinal dynamics of heavy-duty trains;

[0028] Neural network building blocks for neural networks and reward functions for multi-objective optimization of long downhill slopes for heavy-haul trains;

[0029] Iterative optimization module, used to compare the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and use the loss function to iteratively update the neural network parameters;

[0030] Strategy generation module: used to calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are reached, and obtain the optimized control strategy.

[0031] Storage medium for optimizing long downhill slope of multi-locomotive traction heavy-load train, used to store computer software program for optimization calculation,

[0032] The computer program is used to execute an optimization method for realizing a long downhill slope of a multi-locomotive traction heavy-load train.

[0033] It can be seen from the above technical solution that, compared with the prior art, the present invention provides a method, system and storage medium for optimizing a long downhill section of a heavy-load train pulled by multiple locomotives, which has the following beneficial effects: the present invention can optimize the operating curve of a heavy-load train pulled by multiple locomotives in a long downhill section, and realize safe, efficient and smooth operation of the heavy-load train. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0035] Figure 1 It is a flow chart of the optimization method for a multi-locomotive traction heavy-load train going downhill on a long slope disclosed in the present invention;

[0036] Figure 2 The train running state calculation flow chart disclosed in the present invention;

[0037] Figure 3 The invention discloses a block diagram of a long downhill optimization system for multi-locomotive traction heavy-load trains. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0040] Reference Figure 1 As shown, the present invention discloses a long downhill optimization method for a multi-locomotive traction heavy-load train, comprising the following steps:

[0041] S1. Obtain train parameter information, train characteristic parameters and line information;

[0042] S2. Construct a multi-particle model of the longitudinal dynamics of a heavy-load train based on train parameter information;

[0043] S3. Establish a neural network and a multi-objective optimization reward function for a long downhill heavy-haul train based on train characteristic parameters and a multi-mass point model of the longitudinal dynamics of a heavy-haul train;

[0044] S4, calculating the train operation until termination, comparing the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and using the loss function to iteratively update the neural network parameters;

[0045] S5. Calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are met, and obtain the optimized control strategy.

[0046] Furthermore, the train parameter information in S1 includes the weight, length, traction characteristic curve, air brake parameters, etc. of the locomotive and freight car.

[0047] Line information includes line length, ramp information, curve information, tunnel information, electrical phase information, fixed speed limit information, temporary speed limit information, etc.

[0048] The train characteristic parameters include train operation status information, speed limit information, slope information, air brake relief information, action information, whether the off-slope point has been reached, etc.

[0049] Train running status information includes running speed, running distance and running time;

[0050] Speed ​​limit information includes the speed limit ahead, length, and distance to the train;

[0051] Ramp information includes slope and slope length;

[0052] The air brake relief information is a binary parameter;

[0053] Action information includes traction / brake / zero position and related levels;

[0054] Whether the vehicle reaches the exit point is a binary parameter.

[0055] Furthermore, constructing a multi-particle model of the longitudinal dynamics of a heavy-loaded train in S2 includes: constructing a basic resistance model based on traction, electric braking force, air resistance and additional resistance, a coupler force model between locomotive and truck, truck and truck, and an air brake time-lag model; coupling the basic resistance model, the coupler force model and the air brake time-lag model to generate a multi-particle model of the longitudinal dynamics of a heavy-loaded train.

[0056] Furthermore, in the basic resistance model, a single locomotive / truck is regarded as a mass point, and the traction force, electric braking force, air braking force, coupling force of adjacent vehicles, basic resistance and additional resistance are comprehensively considered. The expression is as follows:

[0057]

[0058] In the coupler force model, the coupler buffer devices are divided into QKX100 / 13A and MT-2 / 16. QKX100 / 13A is mainly used between locomotives and freight cars, and MT-2 / 16 is mainly used between freight cars. The coupler force model is calculated based on the dynamic characteristic curve of the coupler buffer device.

[0059] The air brake time lag model takes into account that the air brake command is initiated by the master / slave locomotive and transmitted backward through the train pipe. Therefore, there is a time lag in the air brake application time of each freight car. The expression is as follows:

[0060]

[0061] in, It indicates that the i-th slave locomotive passes through the LOCOTROL system after the master locomotive applies air brakes. delay The time-lag model, It represents the delay of n sampling periods T transmitted through the brake main pipe after the j-th freight car applies air brake in the main control locomotive. sample The time-lag model.

[0062] Furthermore, the multi-mass point model expression of the longitudinal dynamics of heavy-duty train is:

[0063]

[0064] Among them, m iis the mass of the i-th carriage; v is the train speed; μ t,i is the traction output rate, μ t,i ∈[0,1], non-locomotive carriages take 0; F t,i is the train traction force; μ eb,i is the electric brake output rate, μ eb,i ∈[0,1], non-locomotive carriages take 0; F eb,i is the electric braking force of the train; μ fb,i Whether to output air brake, the value is 0 or 1; F fb,i F is the electric braking force of the train; c,i-1 is the coupler force generated by the (i-1) carriage; F c,i is the coupler force generated by the (i+1)th carriage; W 1,i W is the basic resistance encountered by the train during operation; 2,i It is the additional resistance encountered by the train during operation.

[0065] Furthermore, the neural network in S3 includes an input layer, an intermediate layer and an output layer. The feature vector formed by normalizing the train feature parameters in the input layer serves as the input vector of the neural network; the intermediate layer outputs the reward value corresponding to the train traction / braking.

[0066] Specifically, the intermediate layer of the neural network can be constructed as a single layer or multiple layers, and the intermediate layer networks are connected using activation functions, and the activation functions include but are not limited to RELU function, sigmoid function, Tanh function, Softmax function, etc.

[0067] Furthermore, the multi-objective optimization reward function for long downhill runs of heavy-load trains in S3 takes punctuality, safety, and brake device loss as optimization objectives.

[0068] Furthermore, the multi-objective optimization reward function can be expressed as

[0069] η1+η2+η3=1,

[0070] Among them, η1, η2, and η3 are the weights of the three optimization objectives of punctuality, safety, and brake device loss; They are the reward values ​​for the three optimization goals of punctuality, safety, and brake device loss.

[0071] Further, see Figure 2 As shown, the calculation of train operation in S4 includes:

[0072] S41, the neural network inputs the feature vector corresponding to the train entry point;

[0073] S42, obtaining a reward value corresponding to train traction / braking through calculation in the middle layer of the neural network;

[0074] S43, using the heavy-load train longitudinal dynamics multi-particle model and line information to calculate the running state of the train at the next node, determine whether it exceeds the speed limit or reaches the off-slope point, if not, construct a feature vector with the current node train state information, and repeat S42-S43;

[0075] S44. If yes, stop the calculation and distribute the corresponding reward.

[0076] Furthermore, in S41, since the longitudinal dynamic model of the heavy-loaded train adopts a multi-particle model, when the master locomotive enters the slope, the following slave locomotives and freight cars have not yet entered the slope. When calculating the longitudinal force of the entire train, a comprehensive calculation should be made in combination with the line conditions where the trains that have not entered the slope are located.

[0077] In S42, since the heavy-load train with multiple locomotives belongs to the distributed marshaling mode, only the master locomotive / slave locomotive can provide traction and electric braking force, while the freight car can only output air braking. Since the heavy-load train is in a long downhill section, it does not output traction during the optimization process. Therefore, the locomotive braking force F is constructed. b The output range expression is as follows:

[0078] F b ∈[-B p ,0],

[0079] Among them, B p It is the maximum braking force, including electric braking force and air braking force.

[0080] In S44, if the running speed of the train at the next node exceeds the maximum speed limit or is lower than the minimum speed limit, the calculation of the current iteration is terminated and a negative reward (penalty) is given to the current state; if the train reaches the off-slope point at the next node, the calculation of the current iteration is terminated and a positive reward is given to the current state.

[0081] Furthermore, the iterative update in S4 includes: calculating a multi-objective optimization reward function according to the train operation state sequence of each iteration, wherein the multi-objective optimization includes the train operation safety, punctuality and the loss of the braking device, etc.; comparing the reward function value with the output value of the neural network, constructing a loss function, which is used to calculate the difference between the reward value and the output value of the neural network, and the expression is as follows:

[0082] Loss i =(R i -reward i ) 2 ,

[0083] Among them, R iis the output value of the neural network; then according to the chain rule of neural network back propagation, the loss function takes the partial derivative of the previous network output matrix and the current network parameter matrix, and updates the neural network parameters through back propagation iteration, which can be specifically expressed as:

[0084]

[0085] in, are the network parameter matrices of the mth and m-1th times in the nth layer respectively; λ n,i For gain; is the output of the nth layer.

[0086] Furthermore, the preset conditions in S5 include: the Euclidean distance is less than a certain threshold or the maximum number of iterations is reached.

[0087] Specifically, the Euclidean distance represents the Euclidean norm or the L2 norm, which represents the geometric distance before and after the network parameter matrix is ​​updated, and the expression is as follows:

[0088]

[0089] D n =θ n -θ n-1 ,

[0090] Where N is the number of intermediate layers; ||D n ||2 is the 2-norm of the geometric distance of the n-th layer network parameters.

[0091] A multi-locomotive traction heavy-load train long downhill optimization system executes any of the above-mentioned multi-locomotive traction heavy-load train long downhill optimization methods, referring to Figure 3 As shown, it includes an initialization module, a dynamics model building module, a neural network building module, an iterative optimization module and a strategy generation module connected in sequence;

[0092] Initialization module, used to input train parameter information, train characteristic parameters and line information;

[0093] Dynamic model building module, used to build a multi-particle model of the longitudinal dynamics of heavy-duty trains;

[0094] Neural network building blocks for neural networks and reward functions for multi-objective optimization of long downhill slopes for heavy-haul trains;

[0095] Iterative optimization module, used to compare the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and use the loss function to iteratively update the neural network parameters;

[0096] Strategy generation module: used to calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are reached, and obtain the optimized control strategy.

[0097] Storage medium for optimizing long downhill slope of multi-locomotive traction heavy-load train, used to store computer software program for optimization calculation,

[0098] The computer program is used to execute an optimization method for realizing a long downhill slope of a multi-locomotive traction heavy-load train.

[0099] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0100] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing a long downhill slope for a heavy-load train pulled by multiple locomotives, characterized in that: The following steps are involved: S1. Obtain train parameter information, train characteristic parameters and line information; S2. Construct a multi-particle model of the longitudinal dynamics of a heavy-load train based on train parameter information; S3. Establish a neural network and a multi-objective optimization reward function for a long downhill heavy-haul train based on train characteristic parameters and a multi-mass point model of the longitudinal dynamics of a heavy-haul train; S4, calculating the train operation until termination, comparing the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and using the loss function to iteratively update the neural network parameters; S5. Calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are met, and obtain the optimized control strategy.

2. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 1 is characterized in that: The construction of the multi-particle model of the longitudinal dynamics of a heavy-loaded train in S2 includes: constructing a basic resistance model based on traction, electric braking force, air resistance and additional resistance, a coupler force model between locomotive and truck, truck and truck, and an air brake time-lag model; coupling the basic resistance model, the coupler force model and the air brake time-lag model to generate a multi-particle model of the longitudinal dynamics of a heavy-loaded train.

3. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 2 is characterized in that: The multi-mass point model expression of the longitudinal dynamics of heavy-duty train is: Among them, m i is the mass of the i-th carriage; v is the train speed; μ t,i is the traction output rate, μ t,i ∈[0,1], non-locomotive carriages take 0; F t,i is the train traction force; μ eb,i is the electric brake output rate, μ eb,i ∈[0,1], non-locomotive carriages take 0; F eb,i is the electric braking force of the train; μ fb,i Whether to output air brake, the value is 0 or 1; F fb,i F is the electric braking force of the train; c,i-1 is the coupler force generated by the (i-1) carriage; F c,i is the coupler force generated by the (i+1)th carriage; W 1,i W is the basic resistance encountered by the train during operation; 2,i It is the additional resistance encountered by the train during operation.

4. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 1 is characterized in that: The neural network in S3 includes an input layer, an intermediate layer and an output layer. The feature vector formed by normalizing the train feature parameters in the input layer serves as the input vector of the neural network; the intermediate layer outputs the reward value corresponding to the train traction / braking.

5. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 1 is characterized in that: The multi-objective optimization reward function for long downhill slopes of medium- and heavy-load trains in S3 takes punctuality, safety, and brake device loss as optimization objectives.

6. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 4 is characterized in that: The calculation of train operation in S4 includes: S41, the neural network inputs the feature vector corresponding to the train entry point; S42, obtaining a reward value corresponding to train traction / braking through calculation in the middle layer of the neural network; S43, using the heavy-load train longitudinal dynamics multi-particle model and line information to calculate the running state of the train at the next node, determine whether it exceeds the speed limit or reaches the off-slope point, if not, construct a feature vector with the current node train state information, and repeat steps S42-S43; S44. If yes, stop the calculation and distribute the corresponding reward.

7. The long downhill optimization method for a multi-locomotive heavy-load train according to claim 1 is characterized in that: The preset conditions in S5 include: the Euclidean distance is less than a certain threshold or the maximum number of iterations is reached.

8. The optimization system for multi-locomotive traction of heavy-load trains on long downhill slopes is characterized by: The method for optimizing a long downhill slope of a multi-engine traction heavy-load train according to any one of claims 1 to 7 is implemented, comprising an initialization module, a dynamic model construction module, a neural network construction module, an iterative optimization module and a strategy generation module connected in sequence; Initialization module, used to input train parameter information, train characteristic parameters and line information; Dynamic model building module, used to build a multi-particle model of the longitudinal dynamics of heavy-duty trains; Neural network building blocks for neural networks and reward functions for multi-objective optimization of long downhill slopes for heavy-haul trains; Iterative optimization module, used to compare the output value of the multi-objective optimization reward function of the heavy-load train long downhill with the reward value generated by the neural network to construct a loss function, and use the loss function to iteratively update the neural network parameters; Strategy generation module: used to calculate the Euclidean distance of the neural network parameters before and after the update, stop updating when the preset conditions are reached, and obtain the optimized control strategy.

9. A storage medium optimized for long downhill slopes with multiple locomotives pulling heavy-load trains, characterized in that: The storage medium is used to store a computer software program for optimizing calculations, The computer program is used to execute the long downhill optimization method for multi-locomotive traction heavy-load trains as described in any one of claims 1-7.

Images