Rail train traction and braking control method and system

Abstract

The invention provides a rail train traction and brake control method and system, and the method comprises the steps: firstly monitoring the wheel rotation speed and motor torque of a rail train, inputting a state observer with pole configuration to estimate an adhesion coefficient in real time, and combining the creep speed at the current moment to obtain a traction and brake coefficient; calculating the stick-slip slope of the adhesion characteristic curve by using an adhesion coefficient-creep speed empirical formula; the method comprises the following steps: setting a target creep speed by adopting a steepest gradient method, inputting the target creep speed and a current creep speed into a neural network controller, outputting a rail train correction torque, combining the rail train correction torque with a current actual braking torque to obtain a target torque, and controlling a rail train motor through a DTC method so that the actual torque is consistent with the target torque. And the motor torque and the wheel speed are input into the state observer to form closed-loop control. The adhesion coefficient is estimated and optimized in real time through the control method, the train traction and braking efficiency is remarkably improved, and a safer and more efficient solution is provided for modern rail transit.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit train control, and in particular to a rail transit train traction and braking control method and system. Background Art

[0002] With the rapid development of urban rail transit, the operating efficiency and safety of rail trains are becoming increasingly important. Traction and braking are two key dynamic processes in train operation, and their efficiency directly affects the performance, energy consumption and passenger comfort of rail trains. In the actual operation of rail trains, due to factors such as track conditions, weather changes and train load, the adhesion coefficient between the wheelset and the track will fluctuate, resulting in changes in traction and braking force, thereby affecting the acceleration, deceleration movement and operation stability of the train. The traction / braking of the train depends on the contact between the wheel and the track, that is, wheel-rail contact. Low adhesion rail surface conditions and non-optimal traction / braking forces applied to the wheels may cause the wheelset to creep, which may cause the wheels to slip or lock without rapid corrective measures. This slip will cause scratches on the rail surface and tread, reduce the quality of train operation, extend the braking distance, affect the train operation efficiency and even eventually cause safety accidents. It can be seen that for high-speed rail trains, wheel-rail adhesion control during operation is of great practical significance, especially in the complex rail surface environment with low adhesion on long and steep slopes.

[0003] The prior art shows that the adhesion contact state between wheels is very complex and is affected by many practical factors. The "adhesion coefficient" is usually used as a numerical indicator to quantify this state. The adhesion coefficient between wheels and rails is closely related to the creep speed. The adhesion characteristic curve drawn by the relationship between the adhesion coefficient and the creep rate can reflect the adhesion between wheels and rails. The adhesion characteristic curve has a peak point. The left side is the adhesion area with a positive stick-slip slope and the right side is the sliding area with a negative stick-slip slope. Under complex low-adhesion conditions, rail trains are prone to untimely control adjustments, resulting in increased relative sliding between wheels and rails. If the control capability of the braking system is insufficient, it may lead to untimely braking response in emergencies, thereby causing large creeping and reducing the adhesion coefficient. Not only can the adhesion utilization rate not be maximized, but the wheels may also slide excessively, causing serious wear of the wheels and rails and affecting their service life. Therefore, in order to maximize the use of wheel-rail adhesion and prevent rail trains from sliding on the rail surface, the adhesion state of the wheels and rails must be kept near the peak point of the adhesion characteristic curve.

[0004] In the prior art, the research on intelligent control methods for rail vehicles mainly focuses on neural network control or fuzzy control. Fuzzy control has a good processing effect on nonlinear problem systems, but due to the complex changes in road conditions and the difficulty in selecting creep standards, it is difficult to establish fuzzy control rules for the system based on control methods based on model theory. Neural network control has great advantages in robust control and has been used to study vehicle adhesion control, but traditional feedforward neural networks are prone to fall into local minimum points. Summary of the invention

[0005] Purpose of the invention: The first purpose of the present invention is to provide a rail train traction and braking control method for realizing intelligent closed-loop control, and the second purpose is to provide a rail train traction and braking control system.

[0006] Technical solution: A rail train traction and braking control method, comprising the following steps:

[0007] (1) Monitor the wheel speed and motor torque of the rail vehicle and input them into the pole-configured state observer to estimate the adhesion coefficient in real time;

[0008] (2) using the real-time estimated adhesion coefficient obtained in step (1) and the creep speed at the current moment, combined with the adhesion coefficient-creep speed empirical formula, and using the recursive least squares method with a forgetting factor to calculate the stick-slip slope of the adhesion characteristic curve;

[0009] (3) using the stick-slip slope obtained in step (2), the target creep speed is set by adopting the fastest gradient method, the target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm, the rail train correction torque is output, and the correction torque is combined with the actual braking torque of the current rail train to obtain the target torque;

[0010] (4) The rail train motor is controlled by the DTC method so that the actual torque output by the motor is consistent with the target torque obtained in step (3), and the motor torque and wheel speed are input into the state observer of the pole configuration in step (1) to form a closed-loop control.

[0011] Specifically, in step (1), the state equation of the state observer of the pole configuration is as follows:

[0012]

[0013] Where: is the motor angular velocity, for The first derivative of m is the motor output torque, J all is the total moment of inertia of the motor shaft, is the motor output torque after gain, for The first-order derivative of , θ1 and θ2 are the expected characteristic roots;

[0014] The formula for calculating the adhesion coefficient is as follows:

[0015]

[0016] Where: μ is the adhesion coefficient, is the motor output torque after gain, i is the transmission ratio, M is the mass of the track train, g is the gravity coefficient, and r is the wheel radius.

[0017] Specifically, in step (2), the empirical formula of adhesion coefficient-creep velocity is:

[0018]

[0019] Where: μ is the adhesion coefficient, v s is the creep velocity, a, b, c, d are characteristic parameters.

[0020] Specifically, the method for constructing the adhesion coefficient-creep speed empirical formula includes: obtaining several groups of adhesion coefficient and creep speed test data at different speeds through low adhesion braking test experiments, analyzing the adhesion coefficient and creep speed test data using the FASTSIM algorithm and variable friction coefficient model, and then using the iterative reweighted least squares algorithm for linear interpolation fitting to obtain the adhesion coefficient-creep speed empirical formula.

[0021] Specifically, in step (2), the stick-slip slope of the adhesion characteristic curve is calculated using the recursive least squares method with a forgetting factor as follows:

[0022]

[0023] Where K is the stick-slip slope, L and G are intermediate variables, λ is the forgetting factor, t is the number of iterations, μ is the sticking coefficient, and v s is the creep speed.

[0024] Specifically, in step (3), the target creep speed is set as follows:

[0025]

[0026] Where: K is the stick-slip slope, t is the number of iterations, v s is the creep velocity, α and β are hyper parameters.

[0027] Specifically, in step (3), the neural network controller based on the LM algorithm includes an input layer, a first hidden layer, a second hidden layer and an output layer, and the output layer contains 1 neuron, which outputs the correction torque of the rail train.

[0028] Specifically, in step (3), the first hidden layer of the neural network controller based on the LM algorithm contains 8 neurons, and the second hidden layer contains 3 neurons.

[0029] Specifically, in step (4), the DTC method for controlling the rail train motor includes: using a hysteresis loop to adjust the adjustment signal of the control flux and torque, and selecting the voltage space vector in combination with the flux sector to achieve control of the rail train motor.

[0030] The present invention also provides a rail vehicle traction and braking control system, comprising:

[0031] Adhesion coefficient estimation module: used to monitor the wheel speed and motor torque of the rail vehicle and input the pole-configured state observer to estimate the adhesion coefficient in real time;

[0032] Adhesion coefficient optimization module: using the real-time estimated adhesion coefficient and the current creep speed, combined with the adhesion coefficient-creep speed empirical formula, the stick-slip slope of the adhesion characteristic curve is calculated using the recursive least squares method with forgetting factor;

[0033] Neural network controller module: Using the stick-slip slope, the target creep speed is set by the fastest gradient method. The target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm. The rail train correction torque is output and combined with the actual braking torque of the current rail train to obtain the target torque.

[0034] Motor direct torque control module: used to control the rail train motor through the DTC method, so that the actual torque output by the motor is consistent with the target torque, and the motor torque and wheel speed are input into the pole-configured state observer to form a closed-loop control.

[0035] Beneficial effects: Compared with the prior art, the present invention has the following significant effects:

[0036] 1. The rail train traction and braking control method designed by the present invention is applied to the power car of the rail train. By estimating and optimizing the adhesion coefficient in real time, the control of the train wheels is achieved, which significantly improves the traction and braking efficiency of the rail train.

[0037] 2. The present invention uses a three-dimensional characteristic fitting model of adhesion coefficient-creep speed-train speed obtained by fitting experimental data to search for the optimal adhesion point in real time to prevent the wheelset of the rail train from slipping, and the wheelset of the rail train is in the optimal adhesion state when there is no slipping, thereby improving the braking performance of the rail train.

[0038] 3. The present invention adopts a neural network controller based on the LM algorithm to output the correction torque of the track train. When facing long ramps and complex low-adhesion rail surfaces, it can quickly adjust the weights and structural parameters of the network, thereby improving the speed and dynamic effect of adhesion control.

[0039] 4. The present invention adopts an asynchronous motor based on direct torque control, which can effectively track the target torque transmitted by the neural network controller with a very small error, so as to flexibly adjust the motor drive torque to maintain optimal stick-slip, and in the process of optimizing the adhesion adjustment, the current distortion is small and the control is stable.

[0040] 5. The present invention is based on a low-adhesion braking test experiment to obtain multiple sets of experimental data of adhesion coefficient-creep speed at different speeds, and uses the FASTSIM algorithm and the variable friction coefficient model to perform numerical analysis on the measured data, and then uses the iterative reweighted least squares algorithm to perform linear interpolation fitting, so as to obtain the optimal adhesion coefficient of different adhesion characteristic curves at different speeds in real time, that is, to construct a three-dimensional characteristic fitting model of adhesion coefficient-creep speed-train speed, and the data source is highly credible. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the control method of Example 1.

[0042] Figure 2 This is a state observer structure diagram of the pole configuration of Example 1.

[0043] Figure 3 It is the adhesion characteristic curve diagram of Example 1.

[0044] Figure 4 This is a structural diagram of the neural network controller of Example 1.

[0045] Figure 5 It is a schematic diagram of the direct torque control method of embodiment 1.

[0046] Figure 6 It is a schematic diagram of a three-phase full-bridge inverter of Example 1.

[0047] Figure 7 This is the inverter switch equivalent circuit diagram of Example 1.

[0048] Figure 8 This is the sector and voltage space vector diagram of Example 1. DETAILED DESCRIPTION

[0049] The preferred embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0050] Example 1

[0051] The rail train traction and braking control method provided by the present invention is applied to the power carriage of the rail train. For a single power carriage, it is applied to four wheel pairs respectively. To illustrate the detailed scheme of this embodiment, it is necessary to first construct a rail train dynamics model:

[0052] The motor side rotation equation is:

[0053]

[0054] In the formula, J w , J m , J all They are the wheelset moment of inertia, motor moment of inertia, and total motor shaft moment of inertia, T a is the load torque, T m is the motor output torque, i is the transmission ratio, ω m is the motor angular velocity, and t is the time.

[0055] The load torque equation is:

[0056]

[0057] Where r is the wheel radius, F μ is the creep velocity v s Related adhesion.

[0058] The dynamic balance equation of a single train is expressed in matrix form as follows:

[0059]

[0060] Where M v , C v , K v and U v are the mass matrix, damping matrix, stiffness matrix and displacement vector of the rail train, respectively. v represents the external force vector acting on the rail train, For U v The second-order derivative of For U v The first derivative of .

[0061] The adhesion equation is:

[0062]

[0063] Where: μ is the adhesion coefficient, M is the mass of the train, and g is the gravity coefficient.

[0064] The empirical formula of adhesion coefficient-creep velocity is:

[0065]

[0066] Where a, b, c, and d are characteristic parameters.

[0067] The specific parameters of the empirical formula of adhesion coefficient-creep speed are obtained as follows: first, multiple sets of experimental data of adhesion coefficient-creep speed at different speeds are obtained through low-adhesion braking test experiments, and the measured data are numerically analyzed using the FASTSIM algorithm and the variable friction coefficient model proposed by Spiryagin and Polach. Then, the iterative reweighted least squares algorithm is used for linear interpolation fitting, so as to obtain the optimal adhesion coefficient of different adhesion characteristic curves at different speeds in real time, that is, to construct a three-dimensional characteristic fitting model of adhesion coefficient-creep speed-train speed.

[0068] Unless otherwise specified, the above algebraic expressions and definitions of parameters are used in this embodiment.

[0069] Based on the above rail train dynamics model, please refer to Figure 1 As shown, a rail train traction and braking control method provided in this embodiment includes the following steps:

[0070] (1) The wheel speed and motor torque of the rail vehicle are monitored and input into the pole-configured state observer to estimate the adhesion coefficient in real time.

[0071] State space equations for a single train system:

[0072]

[0073] In the formula, u=T m , y = ω m , C =

[10] , ω m The first derivative of T a The first derivative of .

[0074] According to the observability criterion of linear steady-state system, we can know the observability matrix N of the above system and its rank:

[0075]

[0076] Among them, J all is not 0, the rank of the system observation matrix is ​​2, the matrix is ​​full rank, and the above system meets the observability requirements.

[0077] The feedback gain matrix D is loaded on the controlled system, such as Figure 3 As shown. The closed-loop system after loading gain D can be described as:

[0078]

[0079] Where D1 and D2 are gain coefficients. Substitution After finishing, we get:

[0080]

[0081] It is recorded as a closed-loop system ∑G=(A-DC, B, C). The transfer function matrix of the closed-loop system is:

[0082]

[0083] The matrix poles D1 and D2 are configured so that the real part of the eigenvalue of (A-DC) is less than 0. At this time, the adhesion observer converges stably.

[0084] Since the closed-loop characteristic polynomial after loading gain K is:

[0085]

[0086] The desired characteristic polynomial is for:

[0087]

[0088] Where θ1 and θ2 are the expected characteristic roots.

[0089] Comparing the two equations, we can get:

[0090]

[0091] The state equation of the state observer with the pole configuration is finally obtained as follows:

[0092]

[0093] Where: for The first derivative of is the motor output torque after gain, for The first-order derivative of , θ1 and θ2 are the expected characteristic roots.

[0094] See also Figure 2 As shown in the figure, it is the structure diagram of the state observer for pole configuration.

[0095] Then we get:

[0096]

[0097] The final adhesion coefficient is as follows:

[0098]

[0099] Where: μ is the adhesion coefficient, is the motor output torque identified after gain, i is the transmission ratio, M is the mass of the track train, and g is the gravity coefficient.

[0100] (2) Using the real-time estimated adhesion coefficient obtained in step (1) and the creep speed at the current moment, combined with the adhesion coefficient-creep speed empirical formula, the stick-slip slope of the adhesion characteristic curve is calculated using the recursive least squares method with a forgetting factor.

[0101] Please refer to Figure 3 As shown in FIG. 1 , a typical adhesion characteristic curve is shown. In order to quickly track the peak point of the adhesion characteristic curve, it is first necessary to identify the state of the stick-slip slope K. In this embodiment, the Forgetting Factor Recursive Least Squares (FFRLS) algorithm is used for calculation, as shown in the following formula:

[0102]

[0103]

[0104] Where K is the stick-slip slope, L and G are intermediate variables, λ is the forgetting factor, t is the number of iterations, μ is the sticking coefficient, and v s is the creep speed.

[0105] (3) Using the stick-slip slope obtained in step (2), the target creep speed is set by adopting the fastest gradient method, and the target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm. The rail train correction torque is output and combined with the actual braking torque of the current rail train to obtain the target torque.

[0106] After identifying the stick-slip slope K, the target creep speed v is set using the fastest gradient method. s (t+1) is as follows:

[0107]

[0108] Where: K is the stick-slip slope, t is the number of iterations, v s is the creep velocity, α and β are hyper parameters.

[0109] Please refer to Figure 4 As shown, in this embodiment, the neural network controller based on the LM (Levenberg-Marquardt) algorithm includes an input layer, a first hidden layer, a second hidden layer and an output layer. The input layer includes the creep velocity v s (t), target creep velocity v s(t+1), creep error e and error delay, etc. The output layer contains 1 neuron, which outputs the track train correction torque ΔT. The first hidden layer contains 8 neurons, and the second hidden layer contains 3 neurons.

[0110]

[0111] Wherein: m is the number of input nodes of the first hidden layer, n is the number of output nodes of the first hidden layer, and c is any number between 1 and 10. It should be noted that the number 8 is selected as a constraint in this embodiment only as an example and not a limitation. In other actual settings, a larger number limit can be selected. The value range of the constant c is 1 to 10 to provide greater flexibility for the design and optimization of the neural network, ensuring that the scale of the network can be adjusted according to actual needs in different application scenarios.

[0112] According to the structural characteristics of the neural network, the input I of the jth neuron node in the first hidden layer 1_j , output O 1_j They are:

[0113]

[0114] In the formula, z i is the i-th network input, b j is the threshold parameter of the neuron, w ji is the weight between the jth neuron and the i-th network input, and v is the number of neurons in the input layer.

[0115] Similarly, the input I of the qth neuron in the second hidden layer can be obtained 2_q , output O 2_q They are:

[0116]

[0117] In the formula, w qi is the weight between the qth neuron and the jth input of the first hidden layer, b q is the neuron threshold parameter.

[0118] Finally, the output of the output layer is:

[0119] y h =∑(O 2_1 ·e p +O 2_2 ·e i +O 2_3 ·e d )

[0120] The error e(x) between the output of the neural network and the expected value is

[0121] e(x) T =[e1,e2,…,e m ]

[0122] Where m is the dimension of the error vector e(x).

[0123] After the final output is obtained through forward propagation, back propagation is required based on the error e(x) to iteratively update the network's weights and other parameters.

[0124] The Levenberg-Marquardt (LM) algorithm is described below.

[0125] The purpose of continuous iteration of the neural network's weights, thresholds and other structural parameters is to approach the optimal solution. For the optimal solution problem, it can be transformed into a nonlinear equation system:

[0126]

[0127] Expressed in vector form as

[0128] F(x)=0

[0129] Where, F=(f1,f2,…,f z ) T , f i (i=1,2,…,z):R n →R is a continuously differentiable real-valued function, and x represents a one-dimensional vector of dimension u. The number of network weights is l and the number of thresholds is ul, as shown in the following formula:

[0130] x T =[w1,w2,…,w l ,b1,b2,…,b u-l ]

[0131] The LM algorithm can be implemented by following the steps below:

[0132] First, establish the trust region model and solve the iterative step length d of the kth step k In the LM algorithm, the trust region model is

[0133]

[0134] In the formula, Δ k is the current trust region radius, d is the iteration step length, L k (d) is the quadratic approximation model function, and J(x) is the Jacobian matrix of F(x), as shown in the following formula.

[0135]

[0136] Where, the partial derivative in the Jacobian matrix can be obtained by the chain derivative rule.

[0137] Get the iteration step length of step k

[0138] d k =-[J(x k ) T J(x k )+μI] -1 J(x k ) T e(x k )

[0139] Where μ is the correction coefficient and I is the unit matrix.

[0140] Then calculate the drop ratio r k , through the trust region radius Δ k Calculate to update the step size d in real time k range of activities.

[0141] Let f(x k )=F(x k ) T F(x k )

[0142] Then the function ‖F(x)‖ 2 The actual decrease is

[0143] Ared k =f(x k +d k )-f(x k )

[0144] Simultaneous quadratic approximation model function L k The estimated decrease in (d) is

[0145] Pred k =L k (d k )-L k (0)

[0146] Define the drop ratio r k To measure the degree of approximation between the quadratic approximation model and the objective function, as shown in the following formula.

[0147]

[0148] Then, according to the drop ratio r k Adjust the trust region radius

[0149]

[0150] Where, 0 <p1<p2<1,0<γ1<1<γ2。

[0151] Finally, update the parameter μ:

[0152] If r k >0, then x k+1 =x k +d k , k = k + 1, ρ = 2, and

[0153]

[0154] If r k ≤0,x k+1 =x k ,ρ=2·ρ, μ=μ·ρ.

[0155] (4) The rail train motor is controlled by the DTC method so that the actual torque output by the motor is consistent with the target torque obtained in step (3), and the motor torque and wheel speed are input into the state observer of the pole configuration in step (1) to form a closed-loop control.

[0156] The principle of the DTC method is to select the optimal voltage space vector based on the position angle of the flux and the hysteresis adjustment information of the flux and torque, so as to effectively realize the drag control of the motor. Figure 5 As shown, the (DTC) direct torque control method includes main steps such as coordinate transformation, flux hysteresis adjustment, torque hysteresis adjustment, and inverter state switch selection.

[0157] To illustrate the DTC method used in this embodiment, a three-phase asynchronous motor model is first established:

[0158] The voltage equation of the asynchronous motor in the two-phase static coordinate system is as follows:

[0159]

[0160] In the formula, u sα 、u sβ 、u rα 、u rβ are the voltage components of the motor stator and rotor on the α and β axes (V), i sα 、i sβ 、i rα 、i rβ is the current component of the stator and rotor in the α and β axes (A), i A 、i B 、i C is the three-phase current (A), R s , R r are the stator resistance and rotor resistance (Ω), L s , L r , Lm They are stator self-inductance, rotor self-inductance and mutual inductance.

[0161] The flux equation is as follows:

[0162]

[0163] In the formula, ψ sα , sβ , rα , rβ It is the electromagnetic component (Wb) of the stator and rotor in the α and β axes.

[0164] The torque equation is as follows:

[0165]

[0166] Where, T e , T L is the electromagnetic torque and load torque (N·m), n p is the number of pole pairs, J is the moment of inertia on the motor side (kg·m 2 ), α is the angular acceleration (rad / s 2 ).

[0167] Coordinate system conversion (2s / 3s transformation):

[0168] In the three-phase stationary coordinate system, the mathematical model of the asynchronous motor is relatively complex and has a strong coupling relationship. In order to achieve a control effect similar to that of a DC motor, the three-phase stationary coordinate system needs to be converted into a two-phase stationary coordinate system for decoupling control. In order to keep the magnetomotive force unchanged, the current relationship before and after the coordinate transformation is

[0169]

[0170] Among them, i A 、i B 、i C is the three-phase current (A) in the three-phase stationary coordinate system, i α 、i β are the α-axis and β-axis currents (A) in the two-phase coordinate system.

[0171] The direct torque control (DTC) method is described below.

[0172] (1) Stator flux sector

[0173] Magnetic flux observation:

[0174] α and β axis components of stator flux ψ sα , sβ and the flux amplitude |ψ s The calculation model is

[0175]

[0176] Stator flux sector position:

[0177] The sector V where the stator flux is located can be determined by the intermediate variables γ1, γ2, γ3 and the switching coefficient k.

[0178]

[0179] The switching coefficient k is

[0180]

[0181] It can be known that the sector V where the stator flux is located

[0182] V=4k1+2k2+k3,V=1,2,…,6

[0183] Torque hysteresis regulation and flux hysteresis regulation:

[0184] The hysteresis loop is used to adjust the output flux and torque switch signals. The principle is that the stator flux observation value |ψ s |With the given value of magnetic flux|ψ s * The error of | is finally output through the hysteresis regulator to output the flux regulation signal φ, and the motor torque T is different from the target torque T calculated by the neural network controller. * The error between them is finally output as a flux regulation signal τ through the hysteresis regulator. The regulation principle of the hysteresis regulator is as follows

[0185]

[0186] Where ΔT is the torque hysteresis bandwidth.

[0187] (2) Voltage space vector selection:

[0188] The present invention adopts a three-phase full-bridge inverter and sets the IGBT component as the switching device of the inverter. The structure of the inverter is as follows Figure 6 As shown, its equivalent principle is as follows Figure 7 shown.

[0189] Please refer to Figure 7 As shown by S A , S B and S C The circuit has 8 working modes, which are denoted as u i (i=0,1,…,7), where the voltage state corresponding to the working state is called the working voltage state, and the voltage state corresponding to the zero action is called the zero voltage state. The IGBT switching conditions in each state are detailed in Table 1 below, where "1" indicates that the line is connected and "0" indicates that it is not connected.

[0190] Table 1 Inverter switch status

[0191]

[0192] If the symbol u is used v (t) represents the space voltage vector output by the inverter, then

[0193]

[0194] In the formula, u a 、u b and u c They represent the phase voltages of the three-phase stator load windings of the asynchronous motor respectively, and j is an imaginary unit.

[0195] Calculate the voltage space vector in each switching state in turn and obtain Figure 8 The entire voltage space vector diagram is shown.

[0196] Voltage space vector selection:

[0197] In the DTC method, the three states of increasing, unchanged, and decreasing of the flux and torque hysteresis adjustment signals φ and τ are respectively corresponding to the symbols "1", "0", and "-1". According to the increase and decrease needs of φ and τ, combined with the flux sector V, the voltage space vector is selected through the voltage vector selection table to realize the control of the motor. The voltage vector selection table is shown in Table 2 below.

[0198] Table 2 Voltage vector selection table

[0199]

[0200] Example 2

[0201] This embodiment provides a rail train traction and braking control system, including:

[0202] Adhesion coefficient estimation module: used to monitor the wheel speed and motor torque of the rail vehicle and input the pole-configured state observer to estimate the adhesion coefficient in real time;

[0203] Adhesion coefficient optimization module: using the real-time estimated adhesion coefficient and the current creep speed, combined with the adhesion coefficient-creep speed empirical formula, the stick-slip slope of the adhesion characteristic curve is calculated using the recursive least squares method with forgetting factor;

[0204] Neural network controller module: Using the stick-slip slope, the target creep speed is set by the fastest gradient method. The target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm. The rail train correction torque is output and combined with the actual braking torque of the current rail train to obtain the target torque.

[0205] Motor direct torque control module: used to control the rail train motor through the DTC method, so that the actual torque output by the motor is consistent with the target torque, and the motor torque and wheel speed are input into the pole-configured state observer to form a closed-loop control.

[0206] The rail train traction and braking control system provided in this embodiment is applied to the power car of the train. For a single power car, it includes four control systems, which are applied to four wheelsets respectively. Through the coordinated action of the four independent control systems, the adhesion coefficient is estimated and optimized in real time, thereby achieving a significant improvement in the traction and braking efficiency of the rail train.

Claims

1. A rail train traction and braking control method, characterized in that: The following steps are involved: (1) Monitor the wheel speed and motor torque of the rail vehicle and input them into the pole-configured state observer to estimate the adhesion coefficient in real time; (2) using the real-time estimated adhesion coefficient obtained in step (1) and the creep speed at the current moment, combined with the adhesion coefficient-creep speed empirical formula, and using the recursive least squares method with a forgetting factor to calculate the stick-slip slope of the adhesion characteristic curve; (3) using the stick-slip slope obtained in step (2), the target creep speed is set by adopting the fastest gradient method, the target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm, the rail train correction torque is output, and the correction torque is combined with the actual braking torque of the current rail train to obtain the target torque; (4) The rail train motor is controlled by the DTC method so that the actual torque output by the motor is consistent with the target torque obtained in step (3), and the motor torque and wheel speed are input into the state observer of the pole configuration in step (1) to form a closed-loop control.

2. The rail vehicle traction and braking control method according to claim 1, characterized in that: In the step (1), the state equation of the state observer of the pole configuration is as follows: Where: is the motor angular velocity, for The first derivative of m is the motor output torque, J all is the total moment of inertia of the motor shaft, is the motor output torque after gain, for The first-order derivative of , θ1 and θ2 are the expected characteristic roots; The formula for calculating the adhesion coefficient is as follows: Where: μ is the adhesion coefficient, is the motor output torque after gain, i is the transmission ratio, M is the mass of the track train, g is the gravity coefficient, and r is the wheel radius.

3. The rail vehicle traction and braking control method according to claim 1, characterized in that: In the step (2), the empirical formula of adhesion coefficient-creep velocity is: Where: μ is the adhesion coefficient, v s is the creep velocity, a, b, c, d are characteristic parameters.

4. The rail vehicle traction and braking control method according to claim 3, characterized in that: The method for constructing the adhesion coefficient-creep speed empirical formula includes: obtaining several groups of adhesion coefficient and creep speed test data at different speeds through low adhesion braking test experiments, analyzing the adhesion coefficient and creep speed test data using the FASTSIM algorithm and a variable friction coefficient model, and then using an iterative reweighted least squares algorithm for linear interpolation fitting to obtain the adhesion coefficient-creep speed empirical formula.

5. The rail vehicle traction and braking control method according to claim 1, characterized in that: In the step (2), the stick-slip slope of the adhesion characteristic curve is calculated by using the recursive least square method with forgetting factor: Where K is the stick-slip slope, L and G are intermediate variables, λ is the forgetting factor, t is the number of iterations, μ is the sticking coefficient, and v s is the creep speed.

6. The rail vehicle traction and braking control method according to claim 1, characterized in that: In step (3), the target creep speed is set as follows: Where: K is the stick-slip slope, t is the number of iterations, v s is the creep velocity, α and β are hyper parameters.

7. The rail vehicle traction and braking control method according to claim 1, characterized in that: In the step (3), the neural network controller based on the LM algorithm includes an input layer, a first hidden layer, a second hidden layer and an output layer, and the output layer contains 1 neuron and outputs the correction torque of the rail train.

8. The rail vehicle traction and braking control method according to claim 7, characterized in that: In the step (3), the first hidden layer of the neural network controller based on the LM algorithm includes 8 neurons, and the second hidden layer includes 3 neurons.

9. The rail vehicle traction and braking control method according to claim 1, characterized in that: In the step (4), the DTC method for controlling the rail train motor includes: using a hysteresis loop to adjust the adjustment signal for controlling the flux and torque, and selecting a voltage space vector in combination with the flux sector to achieve control of the rail train motor.

10. A rail train traction and braking control system, characterized in that: include: Adhesion coefficient estimation module: used to monitor the wheel speed and motor torque of the rail vehicle and input the pole-configured state observer to estimate the adhesion coefficient in real time; Adhesion coefficient optimization module: using the real-time estimated adhesion coefficient and the current creep speed, combined with the adhesion coefficient-creep speed empirical formula, the stick-slip slope of the adhesion characteristic curve is calculated using the recursive least squares method with forgetting factor; Neural network controller module: Using the stick-slip slope, the target creep speed is set by the fastest gradient method. The target creep speed and the current creep speed of the rail train are input into the neural network controller based on the LM algorithm. The rail train correction torque is output and combined with the actual braking torque of the current rail train to obtain the target torque. Motor direct torque control module: used to control the rail train motor through the DTC method, so that the actual torque output by the motor is consistent with the target torque, and the motor torque and wheel speed are input into the pole-configured state observer to form a closed-loop control.

Images