A method for establishing a neural network model for limiting the maximum traction force of a train

Abstract

A method for establishing a neural network model for limiting the maximum traction force of a train, wherein the input of the neural network model is the train speed and the track state, and the output is the maximum traction force limit value of the train; the train is a train pulled by an electric locomotive. The neural network model is trained and modeled using two rounds of training. The first round of training and modeling method is that the input track state is set to dry, and the maximum traction force limit neural network model is approximated and trained with the empirical formula as the target to obtain the empirical formula approximation model; the second round of training and modeling method is that the collected train speed, track state and corresponding maximum traction force limit value sample data are used to train on the basis of the empirical formula approximation model, and after the training is completed, the model parameters are fixed to obtain the maximum traction force limit neural network model, so that the maximum traction force limit of the train can be changed in real time with the changes in the road conditions of the section, and the train is pulled as much as possible without the wheelset idling.

Description

[0001] This invention patent application is a divisional application. The original application number is 202310475248.2. The application date is April 27, 2023. The name of the invention is a maximum traction control system for a train. Technical Field

[0002] The invention belongs to the technical field of train traction control, and in particular relates to a method for establishing a neural network model for limiting the maximum traction force of a train. Background Art

[0003] Train operation is achieved through the interaction between wheels and rails. Only under the premise of ensuring effective adhesion between wheels and rails can the power of traction motors be further utilized. The wheel-rail adhesion characteristics are not only related to the train itself and the wheel-rail materials, but also to a series of uncertain factors that change with time and space, such as line conditions and rail surface cleanliness. If the traction force is greater than the available adhesion between wheels and rails during locomotive operation, the excess traction force will accelerate the wheels to form idling, the relative sliding speed will increase rapidly, and the available adhesion will decrease rapidly, causing wear and even damage to the wheels and rails, which will not only increase the maintenance and repair costs of railway operations, but also threaten the safe operation of locomotives. Due to the ever-changing conditions of locomotive operation, idling cannot be completely avoided due to changes in driver operation or deterioration of track surface conditions during traction. At present, domestic AC and DC locomotives mainly use a combined correction method for anti-idling and anti-skid control. First, the wheel acceleration is judged. When the acceleration exceeds a certain threshold, it means that the idling and sliding phenomenon is more serious, then the driving torque of the driving wheel is quickly and deeply reduced, that is, the locomotive traction is reduced; if the wheel acceleration does not exceed the threshold, the creep speed is judged. When the creep speed exceeds the threshold, the driving torque is adjusted to a large extent. Otherwise, it is judged to be a normal operating condition. The combined correction method currently used uses two or more single threshold conditions to determine whether idling occurs. When idling does not occur, a comprehensive judgment of the idling risk cannot be achieved; when idling has occurred, a comprehensive judgment of the idling degree cannot be achieved. Summary of the invention

[0004] The purpose of the present invention is to provide a method for establishing a neural network model for limiting the maximum traction force of a train, in view of the defects of the existing technology. The input of the neural network model for limiting the maximum traction force is the train speed and the track state, and the output is the limit value of the maximum traction force of the train. The neural network model for limiting the maximum traction force adopts two rounds of training and modeling. The method for the first round of training and modeling is that the input track state is set to dry, and the neural network model for limiting the maximum traction force adopts the function

[0005]

[0006] The approximation training is performed for the target, and the empirical formula approximation model is obtained, where V is the train speed input by the model, and F μmis the maximum traction limit value output by the model, P μ To calculate the adhesion weight, a1, a2, a3, a4, and a5 are empirical formula parameters for calculating the adhesion coefficient. The train is a train pulled by an electric locomotive.

[0007] The second round of training modeling method is to use the collected train speed, track state and corresponding maximum traction limit value sample data to train on the basis of the empirical formula approximation model. After the training is completed, the model parameters are fixed to obtain the maximum traction limit neural network model. The track state includes dry, wet, and sandy.

[0008] The method for establishing a neural network model for limiting the maximum traction force of a train determines whether the train wheelset is idling based on the creep slip change rate and the creep slip, and determines whether to perform idling traction control on the train maximum traction force limit value output by the maximum traction force limit neural network model according to the idling judgment result; the method for determining whether the train wheelset is idling is that when the idling risk value E is greater than or equal to 1, the train wheelset is idling. The idling risk value E is calculated according to the formula

[0009]

[0010] Calculation is performed, where x1 is the creep rate of change, θ1 is the creep rate of change threshold, x2 is the creep, θ2 is the creep threshold, τ is the nonlinear weighted index, γ1 and γ2 are nonlinear weighted factors, and τ≥2, γ1≥1, and γ2≥1. The creep rate of change x1 and creep x2 are both non-negative values.

[0011] The idling traction control is achieved by controlling the idling traction control ratio θ; the traction control ratio θ is the ratio of the train traction after the idling traction control to the train traction before the idling traction control, and 0≤θ≤1. The idling traction control process is:

[0012] Process I, the process of reducing idling traction, starts when the idling risk value E is greater than or equal to 1 and continues to increase, and ends when the idling risk value E changes from continuously increasing to starting to continuously decrease; in process I, θ is controlled to decrease with a slope d1, and the value of θ at the end of process I is the lowest maintenance value; the lowest maintenance value of θ is not less than 0.

[0013] Process II, the process of maintaining the minimum maintenance value of idling traction, starts from the end of process I and ends when the idling risk value E is less than 1; in process II, the idling risk value E continues to decrease, and θ is controlled to be equal to the minimum maintenance value.

[0014] Process III, the idling traction recovery process, starts from the end of process II and ends when θ increases to 1; in process III, the idling traction control module controls θ to increase with a slope d2 until θ is equal to 1; the decreasing rate of slope d1 is greater than the increasing rate of slope d2.

[0015] In the idling traction control process II, if the idling risk value E changes from continuously decreasing to continuously increasing, then the process returns to process I for idling traction control; in the idling traction control process III, if the idling risk value E increases again to be greater than or equal to 1, then the process returns to process I for idling traction control.

[0016] The method for establishing a train maximum traction force limiting neural network model is used for a train maximum traction force control system. The train maximum traction force control system includes a maximum traction force limiting neural network model, a traction force limiting module and an idling traction force control module. The train traction force output by the train speed controller is controlled by the train maximum traction force limiting value output by the maximum traction force limiting neural network model through the traction force limiting module, and then the idling traction force control module performs idling traction force control on the train traction force, that is, performs train traction force load reduction control.

[0017] The train's maximum traction control system also includes a track surface monitoring module that classifies, identifies and outputs the track status; the track surface monitoring module includes a track surface image acquisition unit and a track surface image recognition unit, and the track surface image recognition unit identifies and processes the real-time track surface image acquired by the track surface image acquisition unit and outputs the current track status; the track status includes three states: dry, wet, and sandy.

[0018] The train maximum traction control system also includes a train speed adjustment processing module. The train speed adjustment processing module periodically collects the train wheel rotation speed, train radar speed and on-board satellite positioning system speed, and calculates the train speed, creep, creep change rate, and train wheel speed change rate. Specifically, the train wheel rotation speed acquisition unit periodically collects the train wheel rotation speed V(h), the train radar speed acquisition unit periodically collects the train radar speed W(h), and the on-board satellite positioning system speed acquisition unit periodically collects the on-board satellite positioning system speed U(k) and positioning status information X(k). The period for collecting the on-board satellite positioning system speed and positioning status information is T U The period of collecting train radar speed and train wheel rotation speed is T V ; T U Greater than T V. The speed adjustment calculation unit adjusts and estimates the wheel / vehicle speed ratio adjustment model parameters and the train radar speed adjustment model parameters according to the vehicle-borne satellite positioning system speed U(k) and the positioning status information X(k), or adjusts and estimates the wheel / vehicle speed ratio adjustment model parameters according to the radar synchronization adjustment speed output by the train radar speed adjustment model. The wheel / vehicle speed ratio adjustment model outputs the train speed, creep, creep change rate, and train wheelset speed change rate. The wheel / vehicle speed ratio adjustment model parameters and the train radar speed adjustment model parameters are adjusted and estimated by an iterative calculation method, and the period of the iterative calculation is the same as the collection period T of the vehicle-borne satellite positioning system speed collection unit. U same.

[0019] The train speed adjustment processing module periodically reads the train wheel rotation speed V(k) and train radar speed W(k) collected at the synchronous collection time point and performs iterative calculations, where k is the current iterative calculation generation.

[0020] During the kth iteration, the train speed adjustment processing module determines whether the vehicle-mounted satellite positioning system speed is valid. If the vehicle-mounted satellite positioning system speed is valid, the current wheel / vehicle speed adjustment coefficient P V (k) According to the formula

[0021]

[0022] The calibration is performed, where U(k) is the most recently acquired vehicle-mounted satellite positioning system speed, and V(k) is the train wheel rotation speed acquired at the time point when U(k) is synchronously acquired.

[0023] When it is judged that the vehicle-mounted satellite positioning system speed is valid, the current radar speed ratio coefficient P W (k) According to the formula

[0024]

[0025] For adjustment, W(k) is the train radar speed collected at the time point of U(k) synchronization collection. When the vehicle-mounted satellite positioning system speed is valid, the radar speed adjustment coefficient P W Equal to P W (k).

[0026] When the vehicle-mounted satellite positioning system speed is judged to be invalid, the current radar speed ratio coefficient P W (k) According to the formula

[0027]

[0028] When the vehicle-mounted satellite positioning system speed is invalid, the radar speed adjustment coefficient P W According to the formula

[0029]

[0030] Calculate, where i=0,1,2,……,m-1, P W (ki) is the radar velocity ratio coefficient for the most recent m times, μ W (ki) is the same as P W (ki) The corresponding radar speed weighting coefficient. Radar synchronous adjustment speed W * (k) According to the formula

[0031]

[0032] Calculate the current wheel / vehicle speed adjustment coefficient according to the formula

[0033]

[0034] The radar speed weighting coefficient satisfies the formula

[0035] The value of each radar speed weighting coefficient satisfies the relationship that the smaller i is, the larger the radar speed weighting coefficient is.

[0036] Wheel / vehicle speed ratio coefficient U V (k) According to the formula

[0037]

[0038] Calculate, where P V (ki) is the speed adjustment coefficient of the most recent m wheels / vehicles when i=0,1,2,...,m-1, μ V (ki) is the same as P V (ki) The corresponding transformation ratio weighting coefficient; m is an integer greater than or equal to 3. The transformation ratio weighting coefficient satisfies the formula

[0039] When taking values ​​of each transformation ratio weighting coefficient, the relationship that the smaller i is, the larger the transformation ratio weighting coefficient is is satisfied.

[0040] Current train speed V C (h) According to the formula

[0041]

[0042] Perform calculations, calculation period and sampling period T V The same. Take the train speed V as the current train speed V C (h).

[0043] Creep x2 according to the formula

[0044]

[0045] Perform calculations.

[0046] The creep rate of change x1 is calculated according to the formula

[0047]

[0048] Calculate, U V (k-1) is the wheel / vehicle speed ratio coefficient obtained from the previous calculation.

[0049] The train wheel speed change rate x3 is calculated according to the formula

[0050]

[0051] Calculation is performed, V(h) is the currently collected train wheel rotation speed, and V(h-1) is the previously collected train wheel rotation speed.

[0052] The τth train wheel rotation speed collection moment before the sampling moment of the vehicle-borne satellite positioning system speed U(k) is the U(k) synchronous collection time point, τ is the delay interval cycle number, and the train wheel rotation speed collected at this point is V(k). Similarly, the τth train radar speed collection moment before the sampling moment of the vehicle-borne satellite positioning system speed U(k) is the U(k) synchronous collection time point, and the synchronous collection time points of the train radar speed W(k) and the train wheel rotation speed V(k) are consistent. The delay interval cycle number τ is the collection period T obtained by converting the time lag value of the vehicle-borne satellite positioning system speed acquisition moment behind the train wheel rotation speed and train radar speed acquisition moments. V When the vehicle-mounted satellite positioning system speed U(k) is invalid, its sampling time still exists, that is, the U(k) synchronous acquisition time point still exists.

[0053] When the relationship between the vehicle speed and the vehicle satellite positioning system speed is judged to be valid for the last m1 consecutive times, the delay interval cycle number τ is calculated, m1 ≥ 10; ε is the acceleration change threshold greater than 0. Specifically, the value of ε can be to Select from the range of values, is the average acceleration of the train at start-up. In the above formula, β(k) is β(ki) when i is equal to 0, which is the most recent calculated rate of change of train acceleration; β(ki) when i is equal to 1, 2, ..., m1-1 is the most recent rate of change of train acceleration m1-1.

[0054] The train acceleration rate is calculated according to the formula

[0055]

[0056] Calculation is performed; wherein α(k) is the most recently collected train acceleration, and α(k-1) is the previously collected train acceleration.

[0057] The train acceleration is measured by an accelerometer. Alternatively, the train acceleration is measured by the formula

[0058]

[0059] Calculate; where U(k-1) is the vehicle-mounted satellite positioning system speed collected the last time U(k) was collected.

[0060] The method for calculating the delay interval period τ is to set the parameter to be optimized as the delay interval period τ * , radar speed proportional coefficient p W * and wheelset speed proportionality factor p V * The delay interval period is τ * When U(ki) is collected at the synchronous collection time point, the train wheel rotation speed is V * (ki), the train radar speed collected at the synchronous collection time point corresponding to U(ki) is W * (ki), that is, the train wheel rotation speed and train radar speed collected at the synchronous collection time point corresponding to U(k) are V * (k), W * (ki), the train wheel rotation speed and train radar speed collected at the synchronous collection time point corresponding to U(k-1) are V * (k-1), W * (k-1), the train wheel rotation speed and train radar speed collected at the synchronous collection time point corresponding to U(k-2) are V * (k-2), W * (k-2), and so on. The minimum optimization objective function is

[0061]

[0062] Take the delay interval period τ that satisfies the optimal value (i.e. Q is the minimum value) Q * is the delay interval period τ; * The value range is greater than 0 and less than 2 / T V integer, p W * and p V * The value range is greater than or equal to 0.8 and less than or equal to 1.2.

[0063] In the train speed adjustment processing module, the sampled train wheel rotation speed is filtered to obtain the collected train wheel rotation speed; the sampled train radar speed is filtered to obtain the collected train radar speed; the sampled vehicle-mounted satellite positioning system speed is filtered to obtain the collected vehicle-mounted satellite positioning system speed. Before collecting the first vehicle-mounted satellite positioning system speed,

[0064]

[0065] Among them, i=1,2,…,m-1.

[0066] The beneficial effects of the present invention are as follows: the maximum traction force limiting neural network model is modeled in a two-round manner, and the established maximum traction force limiting neural network model can take into account the actual running section of the train and the changing running road conditions based on the empirical formula for calculating the adhesion coefficient obtained based on a large amount of experimental data, so that the maximum traction force limit of the train can be changed in real time with the change of the road conditions of the section, and the train is pulled as much as possible without the wheelset idling. When the adhesion condition deteriorates, even if the upper limit is applied, the locomotive traction force (wheel circumferential tangent force) of the wheel axle is still greater than the wheel-rail adhesion force, and the wheelset cannot be prevented from idling. In order to restore the normal traction of the train as soon as possible, the present invention uses a nonlinear mathematical model to calculate the idling risk value E, and integrates the multiple single threshold judgment conditions of the traditional wheelset idling and the weighted judgment conditions when the single threshold conditions are not met into a whole, simplifying the judgment basis, and also when the single threshold conditions are not met, multiple factors are quantified and weighted calculation is performed to achieve a comprehensive judgment of multiple factors, making the idling judgment more comprehensive and accurate. The selection of nonlinear mathematical models can avoid the possibility of misjudgment of weighted judgment conditions when none of the single threshold conditions are met. At the same time, the effect of weighted judgment conditions can be set and adjusted through parameters, and the relative effect of each weighted item can also be set and adjusted through parameters, so that the train wheel set idling judgment method normalized by the nonlinear mathematical model can be applied to different locomotive types and operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a structural schematic diagram of the train maximum traction force control system;

[0068] Figure 2 A flowchart for establishing a train traction force limitation model;

[0069] Figure 3 This is a schematic diagram of the idling traction control module when the train wheelset is idling. Figure 1 ;

[0070] Figure 4This is a schematic diagram of the idling traction control module when the train wheelset is idling. Figure 2 ;

[0071] Figure 5 This is a structural diagram of a train speed adjustment processing module;

[0072] Figure 6 The present invention is a flow chart of a train speed adjustment method;

[0073] Figure 7 It is a schematic diagram of the first-order fitting straight line of the wheel / vehicle speed ratio coefficient;

[0074] Figure 8 A flowchart for calculating the number of delay interval cycles;

[0075] Fig. 9 It is a schematic diagram of the speed acquisition delay of the vehicle-mounted satellite positioning system, the train acceleration, and the train acceleration change rate;

[0076] Fig.10 Schematic diagram of the time points at which the train wheel rotation speed and train radar speed are synchronously collected for the on-board satellite positioning system speed. DETAILED DESCRIPTION

[0077] The present invention is further described below in conjunction with the accompanying drawings. The train is a train pulled by an electric locomotive.

[0078] Figure 1 Schematic diagram of the structure of the train maximum traction control system, including the maximum traction limitation neural network model 10, the traction limitation module 11, the idling traction control module 12, the train speed adjustment processing module 13 and the track surface monitoring module 14. F1 is the train traction output from the train speed controller (i.e., the locomotive speed controller), F μm The maximum traction force limit value of the train is output by the maximum traction force limit neural network model. The traction force limit module is based on F μm The train traction force F1 before the upper limit control is input and the upper limit control is performed on the train traction force, and the output is the train traction force F2 after the upper limit control.

[0079]

[0080] The input of the maximum traction force limitation neural network model is the train speed V and the track state C1, and the output is the train maximum traction force limitation value F μm The method of establishing a train traction force limitation model, that is, establishing a maximum traction force limitation neural network model, is to use two rounds of training and modeling.

[0081] Figure 2The flowchart for establishing the train traction limit model is as follows. The first step is to determine the structure of the neural network model. The track state output by the track surface monitoring module can be a single parameter vector, that is, C1 is a scalar. For the dry state, wet state, and sand state of the identification results, C1 is equal to three constants respectively; for example, let C1 be equal to 2, 1, and 3 for the dry state, wet state, and sand state respectively; at this time, the input parameters of the maximum traction limit neural network model are 2, that is, the input is C1 and V, and the output is F μm ; The maximum traction limit neural network model can select a 2-5-1 3-layer BP neural network structure, or a 2-7-1 3-layer BP neural network structure; When the first round of training modeling is performed, C1 is set to the dry state, that is, C1=2. When the track state is a single parameter vector, for the dry state, wet state, and sand-spreading state of the recognition results, C1 can also be set to correspond to other constant values, for example, C1 is set to correspond to 0, 1, and -1 for the dry state, wet state, and sand-spreading state, respectively. At this time, when the first round of training modeling is performed, C1=0.

[0082] The track state output by the track surface monitoring module can also be a three-parameter vector, that is, C1 = (C 11 , C 12 , C 13 ), 3 parameters C 11 , C 12 , C 13 The recognition results for dry state, wet state, and sandy state are expressed respectively, and the results are all binary values. For example, let the binary values ​​be 1 and 0. When the rail surface state recognition result is dry, C 11 , C 12 , C 13 The values ​​are 1, 0, and 0 respectively; when the rail surface state identification result is wet, C 11 , C 12 , C 13 are 0, 1, and 0 respectively; when the track surface state recognition result is sand, C 11 , C 12 , C 13 are 0, 0, and 1 respectively; or, set the binary value to 1 or -1. When the rail surface state identification result is dry, C 11 , C 12 , C 13 The values ​​are 1, -1, and -1 respectively; when the rail surface state identification result is wet, C 11 , C 12 , C 13 are -1, 1, and -1 respectively; when the track surface state identification result is sand, C 11 , C 12 , C 13The maximum traction force limiting neural network model can select a 4-7-1 3-layer BP neural network structure, or a 4-9-1 3-layer BP neural network structure, or a 4-11-1 3-layer BP neural network structure; when performing the first round of training modeling, C1 is set to a dry state. For example, if the binary value is set to 1 or 0, then C 11 , C 12 , C 13 They are 1, 0, and 0 respectively.

[0083] When the track state output by the track surface monitoring module is a single parameter vector, the maximum traction force limiting neural network model can also select a 3-layer radial basis function neural network with a structure of 2-3-1, 2-4-1, or 2-5-1. When the track state output by the track surface monitoring module is a 3-parameter vector, the maximum traction force limiting neural network model can also select a 3-layer radial basis function neural network with a structure of 4-5-1, 4-6-1, or 4-7-1. The track surface image recognition unit classifies and recognizes the real-time track surface image in dry state, wet state, and sand-sprinkled state and outputs it, which is a conventional image processing technology.

[0084] The second is to determine the empirical formula to be approximated. The empirical formula is

[0085]

[0086] Where V is the train speed input to the model; F μm is the maximum traction limit value output by the model; P μ is the calculated adhesion weight of the train traction locomotive, which is a constant after the locomotive model of the traction train is determined; the expression in brackets is the adhesion coefficient μ j ,Right now

[0087]

[0088] a1, a2, a3, a4, and a5 are empirical formula parameters for calculating the adhesion coefficient, and their values ​​are related to the model of the electric locomotive. For example, domestic electric locomotives of various types take a1 = 0.24, a2 = 12, a3 = 100, a4 = 8, and a5 = 0 respectively; 6K electric locomotives take a1 = 0.189, a2 = 8.86, a3 = 44, a4 ​​= 1, and a5 = 0 respectively; 8G electric locomotives take a1 = 0.28, a2 = 4, a3 = 50, a4 = 6, and a5 = -0.0006 respectively; and so on. The unit of train speed V is km / h; P μ and each traction force F μm The units of F1, F2 and F3 are kN; when necessary, the traction force can also be converted into the corresponding torque.

[0089] The third is to conduct the first round of training modeling. The input track states are dry, wet, and sandy. The input track state is set to dry, and the maximum traction limit neural network model is approximated with the function of formula (2) as the target to obtain the empirical formula approximation model.

[0090] The fourth is to conduct the second round of training modeling. Sample data are collected for train speed, track status and the maximum traction limit value corresponding to the train speed and track status for different types of electric locomotives. When A1, a2, a3, a4, and a5 of the first round of training modeling are empirical formula parameters of various types of domestic electric locomotives, the collected sample data of train speed, track status and the corresponding maximum traction limit value of various types of domestic electric locomotives are used to train on the basis of the empirical formula approximation model. After the training is completed, the model parameters are fixed to obtain the maximum traction limit neural network model of various types of domestic electric locomotives. Similarly, when a1, a2, a3, a4, and a5 of the first round of training modeling are empirical formula parameters of 6K electric locomotives, the collected sample data of train speed, track status and the corresponding maximum traction limit value of 6K electric locomotives are used to train on the basis of the empirical formula approximation model. After the training is completed, the model parameters are fixed to obtain the maximum traction limit neural network model of 6K electric locomotives. When a1, a2, a3, a4, and a5 of the first round of training modeling are empirical formula parameters of 8G electric locomotives, the collected sample data of 8G electric locomotive train speed, track status, and corresponding maximum traction force limit value are used to train on the basis of the empirical formula approximation model. After the training is completed, the model parameters are fixed to obtain the maximum traction force limit neural network model of 8G electric locomotives. Etc. After the training modeling is completed, the maximum traction force limit neural network model inputs the train speed and track status collected in real time by the train speed adjustment processing module and the track surface monitoring module, and outputs the train maximum traction force limit value. The traction force limit module controls the upper limit of the train traction according to the maximum traction force limit value.

[0091] The idling traction control module uses the established nonlinear mathematical model to calculate the idling risk value E. The idling risk value E is calculated according to the formula

[0092]

[0093] Calculate. In formula (4), x1 is the creep rate of change, θ1 is the creep rate of change threshold; X2 is the creep rate, θ2 is the creep rate threshold; τ is the nonlinear weighted index, γ1 and γ2 are nonlinear weighted factors, and τ≥2, γ1≥1, γ2≥1. The creep rate of change x1 and the creep rate x2 are both non-negative values. The idling judgment condition is that when e≥1, it is judged that the train (locomotive) wheelset has idled. Combining formula (4) and the idling judgment condition, the idling judgment logic decomposed is: there are 3 situations (or one of the 3 conditions is met) that can be judged that the train (locomotive) wheelset has idled, namely, ① when the creep rate of change x1 is greater than or equal to the threshold θ1; ② or, when the creep rate x2 is greater than or equal to the threshold θ2; ③ or, when the creep rate of change x1 is less than the threshold θ1 and the creep rate x2 is less than the threshold θ2 and the idling risk value E is greater than or equal to 1. The previous two conditions ①② are single threshold conditions, that is, when a single item satisfies x1≥θ1, or when a single item satisfies x2≥θ2, both satisfy E greater than or equal to 1, that is, the condition for idling judgment is satisfied. Condition ③ is a weighted judgment condition when none of the single threshold conditions are satisfied; the larger the value of τ, the greater the factor of single over-threshold judgment, and the smaller the role of weighted judgment; when τ is large enough, the main body of the idling judgment logic is single threshold conditions ①②, and the role of condition ③ is very small, or almost ineffective; for example, when γ1 and γ2 are both 1 and τ is 200, 0.99 200 =0.134. Even when x1 / θ1 and x2 / θ2 are both equal to 0.99, the idle risk value E is equal to 0.268, which is less than 1 and cannot meet the idle judgment condition. The weighted judgment has almost no effect. When the value of τ is smaller, the weighted effect of condition ③ is more significant. For example, when γ1 and γ2 are both 1 and τ is equal to 2, if x1 / θ1 is equal to 0.8, then when x2 / θ2 is equal to 0.6, E is equal to 1, which has met the idle judgment condition. The nonlinear weighting factors γ1 and γ2 are used to determine the relative effect of the weighted items and do not affect the judgment conditions of each single item exceeding the threshold. The larger the values ​​of γ1 and γ2, the smaller the weighted effect of the corresponding judgment item. Conversely, the smaller the values ​​of γ1 and γ2, the greater the weighted effect of the corresponding judgment item. For example, if γ1 is small and γ2 is large, then in the calculation of the idle risk value E of condition ③, the effect of the item x1 / θ1 in the weighted calculation is greater than that of the item x2 / θ2, but the effect of the single threshold conditions ①② remains unchanged. As long as any one of ①② reaches or exceeds the threshold, the idle judgment condition is still met.

[0094] The idling risk value E or according to the formula

[0095]

[0096] Calculate. In formula (5), x1 is the creep rate of change, θ1 is the creep rate of change threshold; x2 is the creep, θ2 is the creep threshold; x3 is the train wheelset speed change rate, θ3 is the wheelset speed change threshold; τ is the nonlinear weighting index, γ1, γ2, γ3 are nonlinear weighting factors, and τ≥2, γ1≥1, γ2≥1, γ3≥1. The creep rate of change x1, creep rate x2, and train wheelset speed change rate x3 are all non-negative values. The idling judgment condition is that when E≥1, it is judged that the train (locomotive) wheelset has idling. Combining formula (5) and the idling judgment condition, the decomposed train (locomotive) wheelset idling judgment logic is: there are 4 situations (or one of the 4 conditions is met) that can be judged as idling, namely, ① when the creep slip rate x1 is greater than or equal to the threshold value θ1; ② or, when the creep slip rate x2 is greater than or equal to the threshold value θ2; ③ or, when the train wheel set speed change rate x3 is greater than or equal to the threshold value θ3; ④ or, when the creep slip rate x1 is less than the threshold value θ1 and the creep slip rate x2 is less than the threshold value θ2 and the train wheel set speed change rate x3 is less than the threshold value θ3 and the idling risk value E is greater than or equal to 1. The previous 3 conditions ①②③ are single threshold conditions, that is, when the single item satisfies x1≥θ1, or when the single item satisfies x2≥θ2, or when the single item satisfies x3≥θ3, all satisfy E greater than or equal to 1, that is, the idling judgment condition is met. Condition ④ is the weighted judgment condition when none of the single threshold conditions are met. The larger the value of τ is, the greater the factor of single over-threshold judgment is, and the smaller the role of weighted judgment is. When τ is large enough, the main body of the idling judgment logic is single threshold conditions ①②③, and the role of condition ④ is very small, or almost ineffective. For example, when γ1, γ2, and γ3 are all 1 and τ is 200, 0.99 200 =0.134. Even if the ratio of the three judgment items to the corresponding thresholds is 0.99, the idle risk value E is only 0.402, which is less than 1 and cannot meet the idle judgment condition. When the value of τ is small, the weighted effect of condition ④ is greater. For example, when γ1, γ2, and γ3 are all 1 and τ is equal to 2, if x1 / θ1 is equal to 0.8, x2 / θ2 is equal to 0.6, and x3 / θ3 is equal to 0, the idle judgment condition can be met. The nonlinear weighting factors γ1, γ2, and γ3 are used to determine the relative effect between the weighted items and do not affect the judgment condition of each single item exceeding the threshold. The larger the value of γ1, γ2, and γ3, the smaller the weighted effect of the corresponding judgment item; conversely, the smaller the value of γ1, γ2, and γ3, the greater the weighted effect of the corresponding judgment item. For example, if γ1 is small, and γ2 and γ3 are large, then in the calculation of the idling risk value E under condition ④, the effect of the item x1 / θ1 in the weighted calculation is greater than that of items x2 / θ2 and x3 / θ3. However, the effects of the single threshold conditions ①②③ remain unchanged. As long as any one of ①②③ reaches or exceeds the threshold, the idling judgment condition is still met.

[0097] The value range of θ2 is 0.005 to 0.05; the value range of θ1 is 0.0001 / s to 0.005 / s; the value range of θ3 is 3m / s 2 ~30m / s 2 The units of x1, x2, and x3 are the same as those of θ1, θ2, and θ3, respectively.

[0098] The two judgment items in formula (4) or the three judgment items in formula (5) are all power functions with an exponent greater than or equal to 2. The characteristic is that when the value to be judged does not exceed the corresponding threshold, the slope of the curve increases with the increase of the value to be judged, that is, the closer the relevant value is to the corresponding threshold, the greater the impact of its value change on the function item. For example, taking x1 and θ1 as an example, the closer x1 is to θ1, the smaller change of x1 can also cause a larger change in the judgment item. This characteristic of the power function amplifies the effect of the change of the value to be judged (i.e., x1, x2, x3) near the threshold, and is more sensitive near the threshold; conversely, when the value to be judged is much smaller than the threshold, the sensitivity is reduced to minimize the possibility of misjudgment of the weighted judgment condition when the single threshold conditions are not met. In formulas (4) and (5), the larger the value of τ, the greater the change in the slope of the curve when the value to be judged does not exceed the corresponding threshold, that is, it is more sensitive near the threshold and less sensitive away from the threshold.

[0099] The nonlinear mathematical model for calculating the idling risk value E contains creep rate and creep term in both formula (4) and formula (5). Creep is the relative difference between the speed of the train wheelset and the speed of the train. Its value directly reflects how far the train wheelset is from idling, or the degree to which it has already idled. The creep rate is the speed of creep change. This value is related to both the change rate of the train wheelset speed and the change rate of the train speed. The larger its value, the higher the risk of idling. There is also a train wheelset speed change rate term in formula (5). This term is similar to the creep rate change rate. The larger its value, the higher the risk of idling. However, the train wheelset speed change rate has nothing to do with the change of the train speed. The addition of this term can increase the degree of prediction of idling when the train speed is high. When calculating the idling risk value E, you can choose formula (4) or formula (5) as needed; when choosing formula (5), because the effects of the creep change rate and the train wheel speed change rate are similar, they should be taken into consideration when determining the values ​​of γ1 and γ3.

[0100] The nonlinear mathematical model for calculating the idling risk value E, i.e., formula (4) or formula (5), and the corresponding idling judgment conditions, integrates the traditional multiple wheel pair idling single threshold judgment conditions and the weighted judgment conditions when the single threshold conditions are not met into a whole, simplifies the judgment basis, and also quantifies multiple factors and performs weighted calculation when the single threshold conditions are not met, so as to realize the comprehensive judgment of multiple factors and make the idling judgment more comprehensive and accurate. The selection of the nonlinear mathematical model can avoid the possibility of misjudgment of the weighted judgment conditions when the single threshold conditions are not met. At the same time, the effect size of the weighted judgment conditions can be set and adjusted by parameters, and the relative effect size of each weighted item can also be set and adjusted by parameters, so that the train wheel pair idling judgment method normalized by the nonlinear mathematical model can be applied to different locomotive types and operating conditions.

[0101] Figure 3 This is a schematic diagram of the idling traction control module when the train wheelset is idling. Figure 1 The idling traction control ratio θ is the ratio between the train traction output by the idling traction control module and the input train traction, that is, the traction control ratio θ is the ratio between the train traction F3 after idling traction control and the train traction F2 before idling traction control. F3 satisfies the input traction F2.

[0102] F3=θ·F2 0≤θ≤1 (6). Figure 3 Before t1 in , the idling risk value E is less than 1, the train wheelset does not idling, and the idling traction control ratio θ is equal to 1. The idling traction control process of the idling traction control module is:

[0103] Process I, the process of reducing idling traction; starting from when the idling risk value E is greater than or equal to 1 and continues to increase, to when the idling risk value E changes from continuously increasing to starting to decrease, i.e. Figure 3 In process I, the idling traction control module controls θ to decrease at a slope of d1, and the value of θ at the end of process I is the lowest maintenance value. The lowest maintenance value of θ is not less than 0.

[0104] Process II, the process of maintaining the minimum idling traction force; starting from the end of process I, the idling risk value E continues to decrease until the idling risk value E is less than 1, that is, from Figure 3 It starts at time t2 and ends at time t3; in process II, the idling traction control module controls θ to be equal to the minimum maintenance value.

[0105] Process III, idling traction recovery process; starts from the end of process II and ends when θ increases to 1, i.e. Figure 3In process III, the idling traction control module controls θ to increase with a slope d2 until θ is equal to 1. The decreasing rate of the slope d1 is selected between 0.3 / s and 2 / s. For example, when the decreasing rate of the slope d1 is selected as 0.5 / s, θ is reduced by 50% in 1s, which can be reduced from 100% to 50% in 1s, or from 80% to 30% in 1s, etc. The increasing rate of the slope d2 is selected between 0.05 / s and 0.5 / s. For example, when the increasing rate of the slope d2 is selected as 0.2 / s, θ is increased by 20% in 1s, which can be increased from 40% to 60% in 1s, or from 50% to 70% in 1s, etc. When determining d1 and d2, the decreasing rate (absolute value) of the slope d1 is greater than the increasing rate (absolute value) of the slope d2.

[0106] When the idling risk value E increases from less than 1 to greater than or equal to 1, the condition that the idling risk value E is greater than or equal to 1 and continues to increase is satisfied. When θ is equal to 1 and the idling risk value E is continuously less than 1, the idling traction control module does not implement idling traction control.

[0107] Figure 4 This is a schematic diagram of the idling traction control module when the train wheelset is idling. Figure 2 In process II, if the idling risk value E changes from continuously decreasing to continuously increasing, then the process returns to process I to perform idling traction control; Figure 4 In the process, at time t5, the idling risk value E changes from continuously decreasing to continuously increasing, and the idling traction control module immediately returns from process II to process I. In process III, if the idling risk value E increases again to be greater than or equal to 1, it returns to process I to perform idling traction control; Figure 4 At time t6, the idling risk value E increases again to be greater than or equal to 1, and the idling traction control module immediately returns from process III to process I.

[0108] In the currently commonly used combined correction method in China, the unloading strategy of the torque remains unchanged regardless of the degree of idling, and the wheel-rail adhesion state during the unloading process is not considered; first, the unloading depth is not enough, and the idling is not completely suppressed; second, the unloading depth is too large, resulting in the loss of locomotive traction; third, the unloading will be stopped only when the acceleration or creep rate is less than the set threshold, which easily leads to the consequence of excessive unloading depth. The idling traction control module of the present invention performs idling traction control based on the idling risk value that realizes the comprehensive judgment of multiple factors. The degree of unloading of the locomotive traction and the unloading process are controlled by the idling risk value that reflects the wheel-rail adhesion state, which can avoid the situation that the unloading depth is not enough, the idling is not completely suppressed, or the unloading depth is too large, causing the loss of locomotive traction; when the idling risk value changes from increasing to decreasing, the unloading is stopped, which can better avoid the consequences of excessive unloading depth. The nonlinear characteristics of the idling risk value can make the judgment item with a large risk play a relatively more obvious control role.

[0109] Figure 5 The train speed adjustment processing module structure for realizing the train speed adjustment method, or the train speed adjustment system structure diagram. The train wheel rotation speed acquisition unit 101 outputs the collected train wheel rotation speed V(h) (including V(k)) to the speed adjustment calculation unit 104, and the train radar speed acquisition unit 103 outputs the collected train radar speed W(h) (including W(k)) to the speed adjustment calculation unit 104; the vehicle-mounted satellite positioning system speed acquisition unit 102 collects and outputs the vehicle-mounted satellite positioning system speed U(k) and positioning state information X(k) to the speed adjustment calculation unit 104; the speed adjustment calculation unit 104 adjusts and calculates the wheel / vehicle speed ratio adjustment model parameters and the train radar speed adjustment model parameters according to the input information, and outputs the train speed, creep, creep change rate, and train wheel speed change rate. Specifically, the combination switch SW1 in the speed adjustment calculation unit 104 is controlled by the positioning state information X(k) input from terminal 5; when it is determined that the vehicle-mounted satellite positioning system speed is valid according to X(k), terminal 1 of the control combination switch SW1 is connected to terminal 2 and terminal 3, and the parameters of the wheel / vehicle speed ratio adjustment model and the train radar speed adjustment model are adjusted by the vehicle-mounted satellite positioning system speed U(k); terminal 4 is left unconnected, and the radar synchronization adjustment speed W output by the train radar speed adjustment model * (k) is not used at this time, that is, W * (k) does not work at this time. When it is determined that the vehicle-mounted satellite positioning system speed is invalid according to X(k), the terminal 4 of the control SW1 is connected to the terminal 2, and the train radar speed adjustment model recursively deduces the parameters of the train radar speed adjustment model according to the given method. The train radar speed adjustment model adjusts the train radar speed value W(k) at the synchronous acquisition time point in the train radar speed value W(h) to obtain the radar synchronous adjustment speed W* (k), the speed W is adjusted synchronously by the radar * (k) To adjust the parameters of the wheel / vehicle speed ratio adjustment model; Terminals 1 and 3 are suspended, that is, the vehicle-mounted satellite positioning system speed U(k) is not used (or invalid) at this time, and the parameters of the train radar speed adjustment model are not adjusted by external signals. The wheel / vehicle speed ratio adjustment model is adjusted and calculated based on the input train wheel rotation speed V(k) and train radar speed W(k), and the output is the train speed V sent to the maximum traction limit neural network model 10 and the train speed related quantities C2 sent to the idling traction control module 12, wherein when the idling risk value E is calculated according to formula (4), the train speed related quantities C2 include the creep change rate x1 and the creep x2; when the idling risk value E is calculated according to formula (5), the train speed related quantities C2 include the creep change rate x1, the creep x2 and the train wheel speed change rate x3. Figure 5 The combination switch SW1 in FIG. 1 is a schematic switch, which means to control the direction of the signal flow according to X(k). In digital control, the program branching method is usually used to achieve this.

[0110] In the embodiment of the train speed adjustment system, the collection period T of the train wheel rotation speed collection unit is V The acquisition period of the vehicle-mounted satellite positioning system speed acquisition unit is 32ms. U=1s, m is equal to 4. When outputting the train wheel rotation speed V(h) and the train radar speed W(h), the corresponding speed acquisition unit has performed corresponding filtering processing according to the specific situation in the speed sampling and data processing links; for example, if the train wheel rotation speed V(h) is sampled by a pulse speed sensor (encoder), the jitter interference of the pulse edge and the high-frequency interference in the pulse transmission process are filtered out accordingly; if the train wheel rotation speed V(h) and the train radar speed W(h) are directly output as analog or digital quantities, low-pass filtering, smoothing filtering, Kalman filtering, and other filtering methods can be used alone or in combination to filter out high-frequency interference, random interference, white noise interference, etc. The vehicle-mounted satellite positioning system speed acquisition unit includes one or more receiving terminals in the global navigation satellite system GNSS, for example, one or more of the GPS system receiving terminal, the Beidou satellite navigation system receiving terminal, the Galileo satellite navigation system receiving terminal, and the GLONASS system receiving terminal, and also includes a corresponding receiving processing module; the receiving processing module receives information such as the number of satellites being used to solve the position, the ground speed (vehicle-mounted satellite positioning system speed), and whether the positioning state is valid from one or more receiving terminals; or it also includes receiving information such as the longitude, latitude, UTC time, and altitude from one or more receiving terminals, and calculates the vehicle-mounted satellite positioning system speed based on this. The technical means adopted in the train wheel rotation speed acquisition unit, the train radar speed acquisition unit, and the vehicle-mounted satellite positioning system speed acquisition unit are conventional technical means in the field.

[0111] Figure 6 The present invention is a flow chart of a train speed adjustment method for setting the parameters of the wheel / car speed ratio adjustment model and the train radar speed adjustment model in the train speed adjustment system and calculating the wheel / car speed ratio coefficient and various train speed related quantities. The iterative calculation cycle is the same as the acquisition cycle of the speed acquisition unit of the vehicle-mounted satellite positioning system. The specific steps of each iterative calculation are:

[0112] Step 1: Read the kth iteration calculation (equivalent to kT U The vehicle-mounted satellite positioning system data at the sampling time) includes the vehicle-mounted satellite positioning system speed U(k) and positioning state information X(k);

[0113] Step 2, reading the train wheel rotation speed V(k) and train radar speed W(k) collected at the time point of synchronous collection of the vehicle-mounted satellite positioning system speed U(k);

[0114] Step 3, determining whether the vehicle-mounted satellite positioning system speed is valid; if it is determined that the vehicle-mounted satellite positioning system speed is valid, go to step 4; if it is determined that the vehicle-mounted satellite positioning system speed is invalid, go to step 5;

[0115] Step 4: Adjust the model parameters according to the wheel / vehicle speed ratio and radar speed according to U(k), that is, according to the formula

[0116]

[0117] Set the current wheel / vehicle speed adjustment coefficient P V (k) and radar velocity ratio coefficient P W (k); Let radar speed adjustment coefficient P W Equal to P W (k), go to step 6;

[0118] Step 5: Calculate and adjust the radar velocity model parameters, that is, according to the formula

[0119]

[0120] Calculate the current radar speed ratio coefficient P W (k); and

[0121]

[0122] Calculate the radar speed adjustment factor P W According to the formula

[0123]

[0124] Calculate the radar synchronization adjustment speed W * (k); According to W * (k) Adjust the wheel / vehicle speed ratio to adjust the model parameters, that is, according to the formula

[0125]

[0126] Set the current wheel / vehicle speed adjustment coefficient P V (k); Go to step 6;

[0127] Step 6: Calculate the adjustment wheel / vehicle speed ratio coefficient and calculate the speed-related quantities of each train. V (k) There are 2 embodiments; Calculate and adjust the wheel / vehicle speed ratio coefficient U V (k) Example 1, according to formula

[0128]

[0129] Calculate the wheel / vehicle speed ratio coefficient U V (k). Calculate the adjustment wheel / vehicle speed ratio coefficient U V (k) Example 2, for m points (k, P V (k))、(k-1,P V(k-1))、…、(k-m+1,P V (k-m+1)) is used to perform straight line fitting to obtain the first-order fitting straight line of the wheel / vehicle speed adjustment coefficient, and the point (k,U V (k)) value U V (k) is the wheel / vehicle speed ratio coefficient U V (k). Figure 7 Schematic diagram of the first-order fitting straight line of the wheel / vehicle speed adjustment coefficient. Figure 7 In the figure, m is equal to 4, and the four "+" points from left to right are points (k-3, P V (k-3))、(k-2,P V (k-2))、(k-1,P V (k-1))、(k,P V (k)), the point "o" on the first-order fitting straight line of the wheel / vehicle speed adjustment coefficient is the point (k,U V (k)). Figure 7 This is a schematic diagram. The coefficient values ​​of the 4 "+" points are not actual data. For the sake of clarity, the errors are intentionally marked larger, and the slope of the first-order fitting line is also intentionally marked larger.

[0130] The positioning state information X(k) includes information on whether the positioning state is valid positioning or invalid positioning, as well as information on the number of satellites that are currently being used to solve the position. In step 3 of the train speed adjustment method, method 1 for determining whether the speed of the onboard satellite positioning system is valid is that when the positioning state in the positioning state information X(k) is valid positioning, the speed of the onboard satellite positioning system is valid, otherwise, the speed of the onboard satellite positioning system is invalid. Alternatively, method 2 for determining whether the speed of the onboard satellite positioning system is valid is that when the positioning states in the positioning state information X(k) and X(k-1) are both valid positioning, the speed of the onboard satellite positioning system is valid, otherwise, the speed of the onboard satellite positioning system is invalid. Alternatively, method 3 for determining whether the speed of the onboard satellite positioning system is valid is that when the positioning state in the positioning state information X(k) is valid positioning, and the number of satellites that are currently being used to solve the position in the positioning state information X(k) is greater than or equal to δ, the speed of the onboard satellite positioning system is valid, otherwise, the speed of the onboard satellite positioning system is invalid. Alternatively, method 4 for determining whether the speed of the vehicle-mounted satellite positioning system is valid is that when the positioning status in the positioning status information X(k) and X(k-1) are both valid positioning, and the number of satellites being used to solve the position in the positioning status information X(k) and X(k-1) is greater than or equal to δ, the speed of the vehicle-mounted satellite positioning system is valid; otherwise, the speed of the vehicle-mounted satellite positioning system is invalid.

[0131] X(k-1) is the vehicle-mounted satellite positioning system data read at the previous iteration calculation, i.e., at time k-1. In the embodiment, the vehicle-mounted satellite positioning system speed acquisition unit includes a GPS system receiving terminal and a corresponding receiving processing module. Method 3 is used to determine whether the vehicle-mounted satellite positioning system speed is valid, and δ is 5. Generally, δ is required to be greater than or equal to 4.

[0132] P in steps 4-6 V (k), or P when i is equal to 0 V (ki) is the current wheel / vehicle speed adjustment coefficient. P when i is equal to 1, 2, ..., m-1 respectively V (k-1), P V (k-2),…,P V (k-m+1) are the wheel / vehicle speed adjustment coefficients obtained during the first m-1 iterations. W (k), or P when i equals 0 W (ki) is the current radar velocity ratio coefficient. P when i is equal to 1, 2, ..., m respectively W (k-1), P W (k-2),…,P W (jm), are radar velocity ratio coefficients obtained in the first m iterations. W (k), μ W (k-1),…,μ W (k-m+1), which is the same as P W (k), P W (k-1),…,P W The radar velocity weighting coefficient corresponding to (k-m+1) satisfies the formula

[0133] From large to small, μ W (k), μ W (k-1),…,μ W (k-m+1) is used to select the value. For example, when m is equal to 4, μ W (k), μ W (k-1), μ W (k-2), μ W (k-3) is equal to 0.4, 0.3, 0.2, 0.1, or is equal to 0.55, 0.27, 0.13, 0.05, etc.

[0134] Step 6 Calculate the wheel / vehicle speed ratio coefficient U V In Example 1 of (k), μ V (k-1), μ V (k-2),…,μV (k-m+1) is the same as P V (k-1), P V (k-2),…,P V The corresponding transformation ratio weighting coefficient (k-m+1) satisfies the formula

[0135] From large to small, μ V (k-1), μ V (k-2),…,μ V (k-m+1) is used to select the value. For example, when m is equal to 4, μ V (k), μ V (k-1), μ V (k-2), μ V (k-3) is equal to 0.4, 0.3, 0.2, 0.1, or is equal to 0.55, 0.27, 0.13, 0.05, etc.

[0136] In step 6, the train speed related quantities include the train speed V, creep rate x1, creep rate x2, and train wheel speed rate x3. Current train speed V C (h) According to the formula

[0137]

[0138] Perform adjustment calculation, calculation cycle and sampling period T V Same. V(h), V(k), W(h), W(k), U(k), W * (k), V C The unit of (h) is m / s; T V , T U The unit is s. Take the train speed V as the current train speed V C (h); The unit of train speed V is km / h. After converting the unit m / s to km / h, the value of train speed V is equal to V C (h) value is 3.6 times.

[0139] U V (k) reflects the ratio between the wheel speed and the train speed, so the creep slip x2 can be calculated according to the formula

[0140]

[0141] Perform calculations, calculation period and sampling period T U Or, according to the formula

[0142]

[0143] Calculate the current creep degree x2(h), calculation period and sampling period T V Similarly, take the creep degree x2 equal to the current creep degree x2(h).

[0144] The creep rate of change x1 is calculated according to the formula

[0145]

[0146] Perform calculations, calculation period and sampling period T U Same. V (k-1) is the wheel / vehicle speed ratio coefficient obtained by the previous iteration calculation according to the train speed adjustment method. Or, according to the formula

[0147]

[0148] Calculate the current creep rate of change x1, calculation cycle and sampling period T V The same. x2(h-1) is the previous sampling period T V The current creep degree obtained when calculating the creep degree.

[0149] The train wheel speed change rate x3 is calculated according to the formula

[0150]

[0151] Perform calculations, calculation period and sampling period T V V(h-1) is the sampling value before V(h).

[0152] The creep degree x2 and creep degree change rate x1 are calculated by using formula (16) and formula (18). In order to ensure a quick response when calculating the idling risk value E, it is recommended to select formula (5) to calculate the idling risk value E. When the creep degree x2 and creep degree change rate x1 are calculated by using formula (17) and formula (19), formula (4) or formula (5) can be selected as needed to calculate the idling risk value E.

[0153] The τth train wheel rotation speed V(h) acquisition time before the vehicle-borne satellite positioning system speed U(k) sampling time is the vehicle-borne satellite positioning system speed U(k) synchronization acquisition time point, τ is the delay interval cycle number, and the train wheel rotation speed collected at this point is V(k). Similarly, the τth train radar speed acquisition time before the vehicle-borne satellite positioning system speed U(k) sampling time is the U(k) synchronization acquisition time point, and the train radar speed collected at this point is W(k); the train radar speed W(k) and the train wheel rotation speed V(k) have the same synchronization acquisition time point. When the vehicle-borne satellite positioning system speed U(k) is invalid, its sampling time still exists, that is, the U(k) synchronization acquisition time point still exists. Figure 8The flowchart of the method for calculating the number of delay interval cycles in the train speed adjustment system embodiment is the same as the acquisition cycle of the speed acquisition unit of the vehicle-mounted satellite positioning system. The calculation can be performed before or after the iterative calculation of the train speed adjustment method. The specific method is:

[0154] Step ①, get the current time, that is, time k (that is, kT U The train (locomotive) acceleration change rate β(k) at the sampling time);

[0155] Step ②, determine whether the conditions for calculating the number of delay interval cycles are met, and whether the formula is met

[0156] If the relationship between the vehicle speed and the vehicle satellite positioning system speed is valid for the latest m1 times, go to step ③, otherwise exit; m1 is greater than or equal to 10. The acceleration change threshold ε can be selected based on the acceleration capability of the train and combined with experiments. The value of ε can be to Select from the range of values, is the average acceleration of the train when it starts. U For 1s, m1 is equal to 20, and the average acceleration of the train from 0 to 200m can usually reach 0.4m / s 2 , then the value of ε can be selected in the range of 0.4 to 2.4. For example, ε can be taken as 0.75. In formula (21), β(ki) when i is equal to 0 is the train acceleration change rate β(k) at the current moment; β(ki) when i is equal to 1 is the train acceleration change rate obtained when the previous delay interval cycle number is calculated (that is, the wheel / vehicle speed ratio coefficient is iteratively calculated); and so on, β(ki) when i is equal to 1 to m1-1 are the train acceleration change rates obtained when the delay interval cycle number is calculated for the previous m1-1 times. The most recent consecutive m1 times of judging the vehicle-mounted satellite positioning system speed are all valid, which means that according to Figure 6 In the iterative calculation of the train speed adjustment method, in the most recent m1 consecutive iterative calculations, step 3 all determined that the speed of the on-board satellite positioning system was valid.

[0157] Step 3, obtaining the delay interval period τ, the method is, assuming that the parameter to be optimized is the delay interval period τ * , radar speed proportional coefficient p W * and wheelset speed proportionality factor p V * ; τ * The value of is selected within the range that makes the delay interval no more than 2s, that is, τ * Greater than 0 and less than 2 / T V Integer; In the embodiment, T VEqual to 32ms, or 0.032s; 2 / T V is equal to 62.5, so τ * The value range is greater than 0 and less than or equal to 62. W * and p V * The value range of is greater than or equal to 0.8 and less than or equal to 1.2. The parameter p to be optimized W * and p V * Only used in this optimization process. The number of delay interval cycles is τ * When U(ki) is collected at the synchronous collection time point, the train wheel rotation speed is V * (ki), the train radar speed collected at the synchronous collection time point corresponding to U(ki) is W * (ki), the minimum optimization objective function is

[0158]

[0159] Optimization can be done by using genetic algorithms, particle swarm algorithms, and other optimization algorithms to select the delay interval period τ that satisfies the optimal value (minimum value) Q * is the delay interval period τ.

[0160] In step ①, obtain kT U The method of calculating the train acceleration change rate β(k) at the sampling time is as follows:

[0161]

[0162] Calculate, where α(k) is the currently collected train (locomotive) acceleration, and α(k-1) is the last collected train acceleration. In the embodiment, the currently collected train acceleration α(k) is calculated according to the formula

[0163]

[0164] Calculate, where U(k) is the current vehicle-mounted satellite positioning system speed, and U(k-1) is the last vehicle-mounted satellite positioning system speed. The train acceleration α(k) can also be measured and collected using an accelerometer. The unit of α(k) is m / s 2 ; The unit of β(k) is m / s 3 .

[0165] Fig. 9The figure is a schematic diagram of the vehicle-mounted satellite positioning system speed acquisition delay, train acceleration, and train acceleration change rate, where V(t) is the train wheel rotation speed obtained by continuousizing V(h), W(t) is the train radar speed obtained by continuousizing W(h), and U(t) is the satellite positioning system speed obtained by continuousizing U(k). τ The lag time between the vehicle-mounted satellite positioning system speed acquisition time and the train wheel rotation speed acquisition time. Points k-7 to k are the sampling times (k-7)T of the vehicle-mounted satellite positioning system speed. U To kT U ; α(k) and β(k) are the train acceleration and the rate of change of train acceleration respectively.

[0166] Fig.10 The schematic diagram of the synchronous acquisition time point of the train wheel rotation speed and the train radar speed of the vehicle-mounted satellite positioning system speed, where the sampling time (i.e. kT) of U(k) is U ) is the current moment when the train speed adjustment system implements the iterative calculation of the train speed adjustment method. The sampling moments of V(h-τ), V(h-τ+1), ..., V(h-3), V(h-2), V(h-1), V(h), etc. are the sampling moments of the train wheel rotation speed. For example, the moment of V(h) is its sampling moment hT V Due to the influence of ionospheric delay, for the acquisition of train speed (including vehicle-mounted satellite positioning system speed and train radar speed) and train wheel rotation speed at the same time, the acquisition time of the vehicle-mounted satellite positioning system speed lags behind the acquisition time of the train wheel rotation speed and train radar speed. The time lag value is T τ ; The delay interval period τ is relative to the train wheel rotation speed acquisition period T V The number of cycles, that is, the delay interval cycle number τ, is the time lag value of the vehicle-mounted satellite positioning system speed acquisition time lagging behind the train wheel rotation speed and the train radar speed acquisition time, which is converted into the acquisition cycle T V Multiple value. Fig.10 In the example, the sampling time (h-τ) T of V(h-τ) is VThe synchronous acquisition time point of the vehicle-borne satellite positioning system speed U(k) is defined, and the collected train wheel rotation speed V(h-τ) at this point is V(k); specifically, the τth train wheel rotation speed acquisition time (also the train radar speed acquisition time) before the sampling time of the vehicle-borne satellite positioning system speed U(k) is the synchronous acquisition time point of U(k). The acquisition cycle and time of the train radar speed are the same as those of the train wheel rotation speed, and the mutual delay between the two can be ignored. Therefore, the sampling time of the train radar speed W(h-τ), W(h-τ+1), ..., W(h-3), W(h-2), W(h-1), W(h) are the same as the sampling time of the train wheel rotation speed V(h-τ), V(h-τ+1), ..., V(h-3), V(h-2), V(h-1), V(h), respectively. The sampling time (h-τ) of V(h-τ) is (h-τ)T V The sampling time of W(h-τ) is also the synchronous collection time point of the vehicle-mounted satellite positioning system speed U(k). The train radar speed W(h-τ) collected at this point is W(k).

[0167] Similarly, Fig.10 For example, when optimizing the delay interval period τ, if τ * If V(h-1) is equal to 1, the sampling point where V(h-1) is located is its corresponding synchronous acquisition time point, and its V * (k) is equal to V(h-1), W * (k) is equal to W(h-1); if τ * is equal to 2, then the sampling point where V(h-2) is located is its corresponding synchronous acquisition time point, and its V * (k) is equal to V(h-2), W * (k) is equal to W(h-2); and so on. It should be noted that, for example, τ * =1, V * (k) is equal to V(h-1), and V * (k-1) is not V(h-2); In the embodiment, the vehicle-mounted satellite positioning system speed is sampled once, and the train wheel rotation speed is sampled 31.25 times on average. Therefore, if τ * =1, V * (k) is equal to V(h-1), then V * (k-1) may be V(h-32) or V(h-33). When implementing the method for calculating the number of delay interval cycles, the nearest m1·T when collecting various speed data of the train UWithin the time, the train wheelset cannot idle during the train operation; when the train wheelset does not idle, the proportional relationship between the train wheelset speed and the on-board satellite positioning system speed U(k) that reflects the train speed changes little, that is, the proportional coefficient between the actual wheelset speed and the train speed during the train operation is relatively stable, or the wheelset speed proportional coefficient p V * Relatively stable.

[0168] Due to creep, especially the existence of wheelset idling, the train wheelset speed is not consistent with the actual train speed, and when judging whether wheelset idling occurs and calculating creep rate, creep degree and other data, it is necessary to measure the train wheelset speed and train speed separately, and the train wheelset speed cannot be used to replace the train speed. Train speed is commonly measured by radar speed measurement and satellite positioning speed measurement. Satellite positioning speed measurement is to track the train's running speed and position and other information in real time through satellite positioning, and then transmit it to the train control end via satellite for processing, and finally obtain the train speed; satellite positioning speed measurement can overcome the errors caused by wheelset idling and slipping, but the satellite positioning capability is greatly affected by weather and terrain, and speed measurement cannot be achieved 100% of the time; there is a data transmission delay and the transmission delay time is not fixed due to changes in distance and ionosphere conditions, which affects the real-time nature of speed measurement. The radar speed measuring device is generally installed on the bottom of the train, and the radar antenna is at a certain angle to the ground. The radar wave is emitted in the direction of the train. When the train moves relative to the ground, the received radar wave will produce a frequency shift. Based on the radar wavelength, frequency shift, and angle The train speed can be obtained by solving the radar installation height and other data; but the angle Data such as radar installation height may produce time-shift fluctuations, and the road conditions of trains are not consistent. The radar installation height may also change with the road conditions, which may affect the accuracy of radar speed measurement. In the train speed adjustment system for realizing the train speed adjustment method, when the satellite positioning speed measurement is effective, the satellite positioning speed measurement data is used to adjust the wheel / vehicle speed ratio adjustment model parameters and the train radar speed adjustment model parameters; when the satellite positioning speed measurement is invalid, the previously adjusted train radar speed adjustment model parameters are used to calculate the new train radar speed adjustment model parameters according to the given expression, or the first-order fitting straight line method is used to calculate the wheel / vehicle speed ratio adjustment model parameters, and then the wheel / vehicle speed ratio adjustment model parameters are adjusted according to the wheel / vehicle speed ratio adjustment model. The train speed, creep change rate, creep and train wheel speed change rate and other train speed-related quantities are calculated. This method combines the advantages of high satellite positioning speed measurement accuracy, good real-time radar speed measurement and long-term normal operation, and improves the accuracy and reliability of measuring various train speed-related quantities. The train speed adjustment method also determines whether the train is in a speed-changing state. If it is in a speed-changing state, the information obtained by radar speed measurement, satellite positioning speed measurement, and train wheel speed measurement after the train speed change is collected to perform satellite positioning data transmission time, that is, the number of delay interval cycles. Optimize the calculation to obtain accurate real-time satellite positioning data transmission delay time (that is, the number of delay interval cycles), further ensuring the accuracy and reliability of the relevant speed data calculated by the aforementioned train speed adjustment method.

Claims

1. A method for establishing a neural network model for limiting the maximum traction force of a train, characterized in that: The input of the maximum traction force limitation neural network model is the train speed and track state, and the output is the train maximum traction force limit value; the maximum traction force limitation neural network model adopts two rounds of training modeling. The method of the first round of training modeling is that the input track state is set to dry, and the maximum traction force limitation neural network model is based on the function The approximation training is performed for the target, and the empirical formula approximation model is obtained, where V is the train speed input by the model, and F μm is the maximum traction limit value output by the model, P μ To calculate the adhesion weight, a1, a2, a3, a4, and a5 are the empirical formula parameters for calculating the adhesion coefficient; the train is a train pulled by an electric locomotive; The method for the second round of training modeling is to use the collected sample data of train speed, track status and corresponding maximum traction limit value, and conduct training based on the empirical formula approximation model. After the training is completed, the model parameters are fixed to obtain the maximum traction limit neural network model.

2. The method for establishing a neural network model for limiting the maximum traction force of a train according to claim 1, characterized in that: Track conditions include dry, wet, and sandy.

3. The method for establishing a neural network model for limiting the maximum traction force of a train according to claim 1, characterized in that: Whether the train wheelset is idling is determined based on the creep rate of change and the creep degree, and whether to perform idling traction control on the train maximum traction limit value output by the maximum traction limit neural network model according to the idling determination result; the method for determining whether the train wheelset is idling is that when the idling risk value E is greater than or equal to 1, the train wheelset is idling; the idling risk value E is calculated according to the formula Calculation is performed, where x1 is the creep rate of change, θ1 is the creep rate of change threshold; x2 is the creep, θ2 is the creep threshold; τ is the nonlinear weighted index, γ1 and γ2 are nonlinear weighted factors, and τ≥2, γ1≥1, and γ2≥1.

4. The method for establishing a neural network model for limiting the maximum traction force of a train according to claim 3, characterized in that: The idling traction control is achieved by controlling the idling traction control ratio θ; the traction control ratio θ is the ratio of the train traction after the idling traction control to the train traction before the idling traction control, and 0≤θ≤1; the idling traction control process is: Process I, the process of reducing idling traction, starts when the idling risk value E is greater than or equal to 1 and continues to increase, and ends when the idling risk value E changes from continuously increasing to starting to continuously decrease; in process I, θ is controlled to decrease with a slope d1, and the value of θ at the end of process I is the lowest maintenance value; the lowest maintenance value of θ is not less than 0; Process II, the process of maintaining the minimum maintenance value of idling traction, starts from the end of process I and ends when the idling risk value E is less than 1; in process II, the idling risk value E continues to decrease, and θ is controlled to be equal to the minimum maintenance value; Process III, the idling traction recovery process, starts from the end of process II and ends when θ increases to 1; in process III, the idling traction control module controls θ to increase with a slope d2 until θ is equal to 1; the decreasing rate of slope d1 is greater than the increasing rate of slope d2.

5. The method for establishing a neural network model for limiting the maximum traction force of a train according to claim 4, characterized in that: In the idling traction control process II, if the idling risk value E changes from continuously decreasing to continuously increasing, then the process returns to process I for idling traction control; in the idling traction control process III, if the idling risk value E increases again to be greater than or equal to 1, then the process returns to process I for idling traction control.

Images