Train Adaptive Trajectory Tracking Control Method Based on Neural Network and Sliding Mode Control

Abstract

The present invention discloses a train adaptive trajectory tracking control method based on neural network and sliding mode control, which includes the following steps: obtaining the desired speed and desired position trajectory data of the train; the sensors obtain the real-time speed and position data of the train, and compare them with the desired speed and desired position trajectory data to obtain the speed tracking error and the position tracking error; calculating the uncertain parameters during the train operation; using a saturation function to constrain the uncertain parameters; calculating the current control law of the train based on the sliding mode control and the neural network approximation result; the train ATO controller outputs the control law to control the train traction control unit and the train braking control unit to generate traction or braking force and change the train operation state; judging whether the train arrives at the station, if so, the operation ends, if not, continue to execute the above steps. The proposed invention solves the problems of inaccurate train trajectory tracking control and chattering caused by the sliding mode controller.

Description

Technical Field

[0001] The present invention relates to the field of train operation control, and particularly to a train adaptive trajectory tracking control method based on neural network and sliding mode control. Background Art

[0002] Railway transportation plays a very important role in China's comprehensive transportation system. The continuous increase in train operation speed and density, as well as the complexity and changes of operation scenarios, have brought new challenges to train automatic control systems. The operation of high-speed trains has great uncertainty and randomness. For example, external disturbances such as air pressure, temperature, and humidity, as well as the wear of train control devices. The above internal and external unknown disturbances of the system also bring difficulties to the design of the control layer.

[0003] The existing adaptive sliding mode controller designed for high-speed train operation trajectory tracking combines adaptive control and sliding mode control. As a new control strategy, it can effectively improve control accuracy. In the adaptive sliding mode control method, there is generally an assumption that the uncertain parameters are time-invariant. However, in reality, the uncertain model parameters of real high-speed trains are often time-varying. On the other hand, the high-speed train model is complex and there are many uncertain parameters. The process of approximating the uncertain parameters by adjusting the adaptive law is cumbersome. Therefore, in this case, it is difficult for the controller designed by the general mathematical model to achieve high control accuracy. When using the sliding mode technology to control the train to stop, an obvious problem of the sliding mode control is ignored, that is, the chattering phenomenon caused by the discontinuous term in the sliding mode control law will cause the control execution components to oscillate, resulting in device wear or damage. In addition, the existing technology reduces the chattering of the system by introducing a parameter adaptive mechanism. However, the high-speed train model is complex and there are many uncertain parameters. The process of approximating the uncertain parameters by adjusting the adaptive law is cumbersome. Therefore, it is difficult for general controllers to achieve high control accuracy. Summary of the Invention

[0004] In view of the above deficiencies in the prior art, the train adaptive trajectory tracking control method based on neural network and sliding mode control provided by the present invention improves the trajectory tracking control accuracy and comfort of train operation, and solves the chattering problem and inaccurate control problem brought by the sliding mode controller.

[0005] In order to achieve the above invention object, the technical solution adopted by the present invention is: a train adaptive trajectory tracking control method based on neural network and sliding mode control, characterized in that the method includes the following steps:

[0006] S1: Obtain the train desired speed and desired position trajectory data by using the train speed measurement and positioning system;

[0007] S2: Obtain the real-time speed and position data of the train through sensors, and compare them with the desired speed and desired position trajectory data to obtain the speed tracking error and the position tracking error;

[0008] S3: Use the ATO controller to calculate the uncertainty parameters during the train operation according to the obtained speed tracking error and position tracking error;

[0009] S4: Use the saturation function to constrain the uncertainty parameters;

[0010] S5: According to the constrained uncertainty parameters, calculate the actual control law of the train based on the sliding mode control and the approximation result of the RBF neural network;

[0011] S6: Output the control law through the train ATO controller to control the train traction control unit and the train braking control unit, so that they generate traction or braking force to change the train operation state;

[0012] S7: Judge whether the train arrives at the station through the ATP controller. If so, the operation ends. If not, return to step S2.

[0013] The beneficial effects of the above solution are: Through the above technical solution, the combination of the RBF neural network and the sliding mode controller improves the accuracy of trajectory tracking during train operation and solves the chattering phenomenon caused by the discontinuous term in the sliding mode control law.

[0014] Furthermore, S3 includes the following steps:

[0015] S3-1: Determine the input δ of the neural network 1 and δ 2 , and the formula is as follows:

[0016]

[0017] where δ 1 is the position tracking error, δ 2 is the speed tracking error, x r (t) is the ideal position, is the ideal speed, x(t) is the actual position of the train operation, is the actual speed of the train operation, and t is the time;

[0018] Then the formula of the sliding mode surface s(t) is:

[0019]

[0020] where β is the design parameter, is the derivative of the position tracking error;

[0021] S3-2: Use the activation function pj (t) converts the input of the neural network into a new output signal to obtain the radial basis vector p(t) of the neural network. The formula is as follows:

[0022]

[0023] where p(t) is the column vector of p j (t), α is the width vector of the activation function, δ is the input of the network, and δ = [δ 1 , δ 2 T , λ j is the center point vector value, j is the number of neurons in the hidden layer, and exp(·) is the exponential function;

[0024] S3-3: Adjust the weights of the neural network according to the tracking error to achieve online training of the weights output by each radial basis vector and continuously update the network weights. The adaptive law of the network weights is designed as follows:

[0025]

[0026] where M is the mass of the train and ω is the design parameter;

[0027] S3-4: Determine the output θ(t) of the neural network according to the trained network weights and radial basis vectors. The formula is as follows:

[0028] θ(t) = W T (t)·p(t)

[0029] where W T (t) is the weight vector of the neural network;

[0030] S3-5: Obtain the uncertainty parameters during the train operation according to the output of the neural network. The formula is as follows:

[0031]

[0032] where a and b are mechanical resistance coefficients, c is the air resistance coefficient, D x is the external environmental interference, and a, b, c, and D x are all uncertainty parameters.

[0033] The beneficial effect of the above further solution is: obtaining the uncertainty parameters generated during the train operation according to the position tracking error and speed tracking error in the input layer of the RBF neural network, and analyzing and processing them to reduce the influence of interference factors on this solution.

[0034] Furthermore, the constraint formula in S4 is as follows: ​

[0035]

[0036] Among them, sat(·) is the saturation function, and θ * is the disturbance reference value, and is the disturbance estimation value approximated by the RBF neural network.

[0037] The beneficial effect of the above further solution is that by using the saturation function to constrain the disturbance approximation value, the chattering phenomenon caused by the sliding mode controller during train operation is solved, and the safety and comfort of train operation are improved.

[0038] Further, the actual control law formula in S5 is as follows:

[0039]

[0040] Among them, F(t) is the actual control law, Z is the design parameter, V(·) is a continuous function, and μ is the parameter determined by the designer.

[0041] The beneficial effect of the above further solution is that the current control law of the train is obtained through the above formula and used to control the traction control unit and the braking control unit, thereby changing the train operation state. Description of the Drawings

[0042] Figure 1 is the flow chart of the train adaptive trajectory tracking control method based on neural network and sliding mode control.

[0043] Figure 2 is the flow chart of using the ATO controller to calculate the uncertainty parameters during train operation according to the obtained speed tracking error and position tracking error.

[0044] Figure 3 is the architecture diagram of the train adaptive trajectory tracking control based on neural network and sliding mode control. Specific Embodiments

[0045] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0046] As Figure 1 shown, a train adaptive trajectory tracking control method based on neural network and sliding mode control includes the following steps:

[0047] S1: Using the train speed measurement and positioning system to obtain the train desired speed and desired position trajectory data;

[0048] S2: Obtaining the real-time speed and position data of the train through sensors, and comparing them with the desired speed and desired position trajectory data to obtain the speed tracking error and the position tracking error;

[0049] S3: Use the ATO controller to calculate the uncertainty parameters during the train operation based on the obtained speed tracking error and position tracking error;

[0050] S4: Use a saturation function to constrain the uncertainty parameters;

[0051] S5: According to the constrained uncertainty parameters, based on the sliding mode control and the approximation result of the RBF neural network, calculate the actual control law of the train;

[0052] S6: Output the control law through the train ATO controller to control the train traction control unit and the train braking control unit, so that they generate traction or braking force to change the train operation state;

[0053] S7: Use the ATP controller to judge whether the train arrives at the station. If so, the operation ends; if not, return to step S2.

[0054] As Figure 2 shown, S3 includes the following steps:

[0055] S3-1: Determine the input δ 1 and δ 2 of the neural network, and the formula is as follows:

[0056]

[0057] where δ 1 is the position tracking error, δ 2 is the speed tracking error, x r (t) is the ideal position, is the ideal speed, x(t) is the actual position of the train operation, is the actual speed of the train operation, and t is the time;

[0058] Then the formula of the sliding mode surface s(t) is:

[0059]

[0060] where β is the design parameter, is the derivative of the position tracking error;

[0061] S3-2: Use the activation function p j (t) to convert the input of the neural network into a new output signal to obtain the radial basis vector p(t) of the neural network, and the formula is as follows:

[0062]

[0063] where p(t) is p jThe column vectors of (t), α is the width vector of the activation function, δ is the input of the network, δ = [δ 1 , δ 2 T , λ j is the center point vector value, j is the number of neurons in the hidden layer, exp(·) is the exponential function;

[0064] S3-3: Adjust the neural network weights according to the tracking error to realize the online training of the weights output by each radial basis vector, and continuously update the network weights. The adaptive law of the network weights is designed as follows:

[0065]

[0066] where M is the mass of the train and ω is the design parameter;

[0067] S3-4: Determine the output θ(t) of the neural network according to the trained network weights and radial basis vectors. The formula is as follows:

[0068] θ(t) = W T (t)·p(t)

[0069] where W T (t) is the weight vector of the neural network;

[0070] S3-5: Obtain the uncertainty parameters during the train operation according to the output of the neural network. The formula is as follows:

[0071]

[0072] where a and b are the mechanical resistance coefficients, c is the air resistance coefficient, D x is the external environmental interference, and a, b, c and D x are all uncertainty parameters. The external environmental interference D x ​It includes additional resistance and unknown disturbances. The additional resistance of the train is the resistance received by the train when passing through special sections. For example, the additional ramp resistance received when passing through a ramp, the additional curve resistance received when passing through a curve, and the additional tunnel resistance received when passing through a tunnel. It can be seen from this that the additional resistance of the train mainly depends on the operating line conditions of the high-speed train. The basic running resistance of the train consists of mechanical resistance and air resistance. Among them, the mechanical resistance is mainly the sliding resistance between the wheels and the rails. The mechanical resistance coefficients a and b depend on the relevant components of the train and the track. Due to the complex structure of these factors, it is difficult to quantitatively calculate them in practice. Therefore, empirical formulas are often obtained through a large number of experiments to calculate the additional running resistance of the train. The air resistance is generated by the friction between the surface of the high-speed train and the air during operation, as well as the compression of the air at the head of the high-speed train and the generation of vortices at the tail of the air. Most of the running resistance of the high-speed train is air resistance. The air resistance coefficient c is often also obtained through a large number of experiments. At the same time, factors such as temperature and air density will also greatly affect the air resistance coefficient.

[0073] In S4, there is a reference value θ for disturbances in engineering. * Considering the safety and comfort of train operation, the approximation value of the disturbance is constrained by a saturation function, and its constraint formula is as follows:

[0074]

[0075] Among them, sat(·) is the saturation function, θ * is the reference value of the disturbance, is the estimated value of the disturbance approximated by the RBF neural network.

[0076] The actual control law formula in S5 is as follows:

[0077]

[0078] Among them, F(t) is the actual control law, Z is the design parameter, V(·) is a continuous function, μ is the parameter determined by the designer.

[0079] In the control architecture of the present invention, as Figure 3 shown, the expected position and speed trajectory data calculated according to the train characteristics, line characteristics and other parameters of the ideal train model; the real-time running position and speed of the train obtained by the sensor; the input signal of the ATO controller, which is generated by the central control unit CCU and transmitted to the traction control unit TCU and the braking control unit BCU to control the train operation; in the RBF neural network, the input position and speed error is approximated to the uncertain disturbance to obtain the estimated value; the adaptive law: the weight vector of the RBF neural network is adjusted by designing the adaptive law.

[0080] In an embodiment of the present invention, the train obtains the desired speed and desired position trajectory data through a speed measurement and positioning system. The sensor obtains the real-time speed and position data of the train, compares them with the desired speed and position trajectory data to obtain the speed tracking error and position tracking error. The ATO controller calculates the uncertainty parameters during the train operation according to the obtained tracking errors, and at the same time uses a saturation function to calculate the uncertainty parameters. Based on the uncertainty parameters, the ATO controller calculates the current control law of the train and outputs the control law according to the results of sliding mode control and neural network approximation, controls the train traction control unit and the train braking control unit to generate traction force or braking force, so as to change the running state of the train. At the same time, the ATP controller determines whether the train arrives at the station. If so, the control ends. If not, the above steps are continued until the train arrives at the station.

[0081] The present invention designs an adaptive law to approximate the weights of the neural network, and then combines it with sliding mode control to design a speed and position trajectory tracking controller for high-speed trains, realizing precise tracking of the desired speed and position trajectory. At the same time, the present invention is a model-free sliding mode controller based on an adaptive neural network that only requires input or output data drive, does not rely on prior knowledge of high-speed trains and their operating environments, and has strong engineering application prospects.

[0082] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention according to the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the invention.

Claims

1. A train adaptive trajectory tracking control method based on neural network and sliding mode control, characterized in that, the method comprises the following steps: S1: Obtain the expected speed and expected position trajectory data of the train by using the train speed measurement and positioning system; S2: Obtain the real-time speed and position data of the train through sensors, and compare them with the expected speed and expected position trajectory data to obtain the speed tracking error and position tracking error; S3: Use the ATO controller to calculate the uncertainty parameters during the train operation according to the obtained speed tracking error and position tracking error; S4: Use a saturation function to constrain the uncertainty parameters; S5: Calculate the actual control law of the train based on the constrained uncertainty parameters, the sliding mode control and the RBF neural network approximation result; S6: Output the control law through the train ATO controller to control the train traction control unit and the train braking control unit, so that they generate traction or braking force to change the train operation state; S7: Judge whether the train arrives at the station through the ATP controller. If so, the operation ends; if not, return to step S2; The S3 includes the following steps: S3-1: Determine the input δ of the neural network 1 and δ 2 , as shown in the following formula: Among them, δ 1 is the position tracking error, and δ 2 is the speed tracking error. x r (t) is the ideal position, is the ideal speed, x(t) is the actual position of the train operation, is the actual speed of the train operation, and t is the time; Then the formula of the sliding mode surface s(t) is: where β is a design parameter, is the derivative of the position tracking error; S3-2: Use the activation function p j (t) to convert the input of the neural network into a new output signal, obtaining the radial basis vector p(t) of the neural network, as shown in the following formula: Among them, p(t) is the column vector of p j (t), α is the width vector of the activation function, δ is the input of the network, δ = [δ 1 , δ 2 T , λ j is the center point vector value, j is the number of neurons in the hidden layer, exp(·) is the exponential function;​ S3-3: Adjust the weights of the neural network according to the tracking error to realize the online training of the weights output by each radial basis vector, continuously update the network weights, and the adaptation law of the network weights The formula is designed as follows: where M is the train mass and ω is the design parameter; S3-4: Determine the output θ(t) of the neural network according to the trained network weights and radial basis vectors, and the formula is as follows: θ(t) = W T (t)·p(t) Among them, W T (t) is the weight vector of the neural network; S3-5: Obtain the uncertainty parameters during the train operation according to the output of the neural network, and the formula is as follows: where a and b are mechanical resistance coefficients, c is the air resistance coefficient, and D x is the external environmental interference, and a, b, c, and D x are all uncertain parameters.

2. The train adaptive trajectory tracking control method based on neural network and sliding mode control according to claim 1, characterized in that, the constraint formula in the S4 is as follows: where sat(·) is the saturation function, and θ * is the disturbance reference value, and is the disturbance estimation value approximated by the RBF neural network.

3. The train adaptive trajectory tracking control method based on neural network and sliding mode control according to claim 2, characterized in that, the formula of the actual control law in the S5 is as follows: Among them, F(t) is the actual control law, Z is the design parameter, V(·) is a continuous function, and μ is the parameter determined by the designer.

Images