CN106707765A - Running-tracking, real-time optimization control method for high speed train - Google Patents

Abstract

The invention discloses a running-tracking, real-time optimization control method for a high speed train. The method comprises the following steps: collecting a practical running data of the high speed train; establishing an offline ANFIS model of the high speed train and designing a prediction controller on the basis of the offline ANFIS model; starting an online model regulation strategy, returning a practical running data, adopting Kalman filtering and BP gradient descending algorithm for optimizing the ANFIS model in real time, thereby regulating the parameter of the prediction controller on line, eliminating the influence of the change in characteristics on the running of the high speed train and realizing the real-time optimization control on the running process of the high speed train, when the tracking performance is poor under the effect of the change in object characteristics or environment and an absolute value|er|of the feedback velocity error exceeds a preset error threshold value x. According to the invention, the running performance of the high speed train can be guaranteed and a beneficial technical support is supplied for an automatic pilot system of the high speed train. The invention is suitable for the technical field of the real-time optimization control for the running tracking of the high speed train.

Description

High-speed motor train unit tracking operation real-time optimization control method

Technical Field

The invention relates to a modeling and operation optimization control method for multiple working conditions in the operation process of a high-speed motor train unit, and belongs to the technical field of tracking operation real-time optimization control of the high-speed motor train unit.

Background

The high-speed motor train unit has the advantages of large passenger capacity, safety, comfort, rapidness, energy conservation, environmental protection, all-weather transportation and the like, is regarded by all countries, and becomes the first choice for developing advanced traffic systems in many countries. With the development of economy and society in China, the large-scale rapid development of high-speed railways is urgently needed, and a train operation control system with safety, punctuality and energy conservation as marks is built. In a rail transit system, a train operation control system is a core system for ensuring that a train can safely operate and improving the operation efficiency, and the quality of a control strategy directly influences the railway transportation capacity. The high-speed motor train unit has high running speed and long distance, the running process of the high-speed motor train unit has high sensitivity to relevant influence factors, and the influence is more and more complex than that of the traditional railway. The existing high-speed motor train unit operation control is a manual operation control mode based on actual operation conditions under the guidance of an automatic train protection system (ATP) by a driver, and the operation performance of the motor train unit is closely related to the operation experience of the driver and the degree of response to faults. Therefore, in order to guarantee high-speed safe operation of the motor train unit in real time, a high-speed motor train unit operation process model needs to be established and the operation process of the high-speed motor train unit needs to be optimized and controlled in real time.

A traditional modeling method for the running process of the high-speed motor train unit generally adopts a description method based on traction calculation and an operation resistance empirical model, but the method cannot completely depict the complex and changeable dynamic behaviors of the motor train unit. In order to more accurately restore the dynamic operation process of the high-speed motor train unit, data-driven modeling gradually becomes a research hotspot for modeling the high-speed motor train unit. For the tracking control of the train running process, a PID control method is more classical. The related scholars design a fuzzy PID method to control the track of the speed of the agile train, and because PID control has no self-adaptive capacity, the fuzzy PID method is only suitable for subway systems with stable environment and low speed. In order to solve the problem, a learner adopts a robust self-adaptive control method to realize the tracking control of the speed and the position of the high-speed motor train unit; however, parameter decomposition and control law design of adaptive robust control need a large amount of calculation, and the actual running environment condition of the high-speed motor train unit cannot be well solved. Considering that the generalized predictive control method can effectively overcome the uncertainty and nonlinearity of the process, can conveniently process various constraints of the process controlled variables, is suitable for a complex uncertain system, and realizes the high-precision control of the speed and the displacement of the high-speed motor train unit in the process of the research of more scholars at present. However, the control methods do not have real-time optimization control performance, and the influence caused by some uncertain factors in the running process of the high-speed motor train unit is difficult to eliminate in time.

Disclosure of Invention

The invention aims to design a tracking operation real-time optimization control system of a high-speed motor train unit to ensure safe, punctual and efficient automatic operation, adjust model parameters on line and improve the safety and punctuality of the operation process of the high-speed motor train unit aiming at the characteristics of complex operation environment, frequent working condition change, multiple factors influencing the operation performance and the like of the high-speed motor train unit.

The technical scheme of the invention is as follows:

a real-time optimization control method for a high-speed motor train unit comprises the steps of establishing an ANFIS (artificial neural network interface) model of the high-speed motor train unit by collecting actual operation data of the high-speed motor train unit, designing a corresponding prediction controller based on the ANFIS model, collecting predicted output speed and expected speed in real time, and analyzing an operation speed error; when the high-speed motor train unit is influenced by uncertain factors such as unknown environment or motor train unit characteristic change and the like on tracking control, the Kalman filtering algorithm and the BP gradient descent method are combined to adjust model parameters in real time so as to adapt to the following high-precision tracking control of the high-speed motor train unit, eliminate the influence of characteristic change on the operation of the motor train unit and ensure the operation performance of the high-speed motor train unit.

The high-speed motor train unit off-line ANFIS model comprises the following steps:

suppose that m high-speed motor train unit operation process data points { X }₁,…,X_i,…,X_mIn which X is_i＝[v_i(k-1),u_i(k-1),v_i(k)]Data point X_iThe density index of (A) is defined as

In the formula,_athe radius of the effective neighborhood of the set clustering center is a positive number; selecting the highest value of the density indexObtaining a first cluster center

Each data point X_iThe density index of (A) is corrected by the following formula

Wherein_bIs one is greater than_aA positive number of; obviously, close to the first cluster center c₁The density indicator of the data points of (a) will be significantly reduced, thus making it less likely that these points will be selected as the next cluster center; after the density index of each data point is corrected, the next clustering center c is selected₂And correcting all density indexes of the data points again, and repeating the process until the density indexes of the data points are corrected againGet the last cluster center c_nTherefore, the number of the clustering centers of the data in the running process of the high-speed motor train unit is n;

obtaining n linear models by adopting minimum variance estimation on n rule postresults; the n linear models are fused to obtain the following offline ANFIS model:

wherein,is the normalized value of all rule fitness.

The real-time optimization control method for the high-speed motor train unit comprises the following steps of:

the running process model (7) of the high-speed motor train unit can be described in the following form

In the formula,andis z^-1A polynomial of (1-z) (. DELTA. -)^-1；

Wherein the parametersAndobtained by a modeling process.

To obtain the optimal control scalar, the performance indicator function can be designed as

Wherein, L, H, G andis the introduced parameter matrix of the loss-of-image equation,is a weighting coefficient matrix;

minimize performance index (i.e.) Obtain the optimal control increment of

According to the real-time optimization control method for the high-speed motor train unit, when the high-speed motor train unit is influenced by uncertain factors such as unknown environment or motor train unit characteristic change and the like on tracking control, an online model adjusting strategy is started to adapt to the following high-precision tracking control of the high-speed motor train unit, the model parameters are adjusted in real time by combining a Kalman filtering algorithm and a BP gradient descent method, and the specific optimization steps can be represented as follows:

step 1, designing a generalized prediction controller based on an ANFIS model of the high-speed motor train unit, and calculating a speed feedback error e at the time t_r(t)＝y(t)-y_r(t)；

Step 2, executing the tracking operation control of the high-speed motor train unit;

step 3, judging whether the time T reaches the total time T or not_tIf the current time reaches the preset time, ending the process, otherwise entering the next time t +1, and turning to Step 4;

at the moment of Step 4, t +1, the feedback error e of the previous moment is calculated_r(t) comparing with a set error threshold χ, if | e_r(t) | is more than or equal to χ, the model mismatch is indicated, the model mismatch is not suitable for the control condition at that time, and the Step5 is required to carry out online adjustment on the model parameters; if the feedback error | e_r(t) | < χ, indicating that the tracking precision is higher, directly returning to Step 1 without optimizing the model parameters;

and Step5, combining a Kalman filtering algorithm and a BP gradient descent method, adjusting parameters of a front part and a back part of the ANFIS model on line, substituting the optimized model into the generalized predictive controller, and returning to Step 1.

The real-time optimization control method for the high-speed motor train unit comprises the Step5, a Kalman filtering algorithm and a BP gradient descent method are combined, and the method for adjusting parameters of a front piece and a back piece of the ANFIS model on line comprises the following steps:

on the basis of the established model (8), firstly, the following Kalman filtering algorithm is adopted to adjust the parameters of the back part

In the formula,forgetting factor 0 < lambda for the back-piece parameters obtained during modeling_KF1A positive number close to 1 (in this context. lambda.) is generally chosen_KF＝0.9995)；P_KF(k)＝q_KFI∈R^2n×2n，q_KFIs a large positive number, usually taken as 10⁴～10¹⁰(in the invention q)_KF＝10⁶)；

Secondly, optimizing the front-part parameter c in real time by adopting a BP gradient descent method_ijAnd σ_ij；

Calculating an error index function of

Wherein the k-th data points y (k) and y_r(k) Respectively representing the actual output speed and the expected output speed of the model transmitted by the control part at the moment of mismatch t, and so on; the front-part parameter optimization algorithm is as follows:

and after the ANFIS model parameters are corrected, substituting the corrected ANFIS model parameters into the generalized predictive controller for recalculation to obtain corresponding control force to implement real-time optimization control on the tracking operation of the high-speed motor train unit.

Compared with the prior art, the high-speed motor train unit has the beneficial effect that the high-speed motor train unit is a complex nonlinear system which operates in a changeable environment. In order to improve the running performance of the high-speed motor train unit, an effective running process controller needs to be designed to accurately control the high-speed motor train unit. The existing control methods for high-speed motor train units designed by researchers do not have a real-time optimization function, and the situation that the tracking performance is poor due to the characteristics of the motor train units or environmental changes is difficult to process, so that the running performance of the high-speed motor train units cannot be guaranteed. The invention provides a novel tracking operation real-time optimization control method based on the characteristics of complex operation environment, frequent working condition change, multiple factors influencing the operation performance and the like of a high-speed motor train unit. Firstly, an offline ANFIS model in the running process of the high-speed motor train unit is established, and a corresponding generalized predictive controller is designed. When the tracking performance of the motor train unit is deteriorated due to the characteristics or environmental changes of the motor train unit, an online adjustment strategy is started, and online adjustment is performed on the ANFIS model of the high-speed motor train unit operation by adopting a Kalman filtering and BP gradient descent method, so that the parameters of the prediction controller are adjusted, the real-time optimization control of the motor train unit tracking operation is realized, and the operation safety and the operating punctuality are improved.

The method is suitable for the real-time optimization control of the tracking operation of the high-speed motor train unit.

Drawings

FIG. 1 is a structural block diagram of a real-time optimization control system of a high-speed motor train unit;

FIG. 2 shows stress conditions of a high-speed motor train unit in the running process;

FIG. 3 is a real-time optimization control flow of the operation process of the high-speed motor train unit;

FIG. 4 is an output error profile of inspection data;

FIG. 5 is a characteristic change speed tracking curve;

FIG. 6 is a characteristic change speed tracking error curve;

FIG. 7 is a characteristic change traction/braking force curve;

FIG. 8 is a characteristic change acceleration curve;

FIG. 9 is a parameter optimization process for a property change ANFIS model;

Detailed Description

The present invention will be described in detail with reference to specific examples.

The method comprises the steps of collecting actual operation data of the high-speed motor train unit, analyzing a tracking operation control mechanism of the motor train unit, establishing an offline ANFIS model of the operation process of the high-speed motor train unit by combining a traction/braking characteristic curve and the actual operation data of the motor train unit, and designing a generalized predictive control algorithm based on the ANFIS model to realize the tracking operation process control of the motor train unit; and when the tracking performance is deteriorated due to the object characteristics or environmental changes, starting a model online adjustment strategy, and optimizing the ANFIS model in real time by adopting Kalman filtering and BP gradient descent algorithm so as to adjust the parameters of the prediction controller online and realize the real-time optimization control of the running process of the high-speed motor train unit.

The method for modeling the operation process of the high-speed train based on the ANFIS comprises the following steps:

1. analyzing the real-time optimization control principle of the high-speed motor train unit:

fig. 1 illustrates a real-time optimization control system structure for the operation of the motor train unit based on the ANFIS model and the generalized predictive control. The method comprises the steps of establishing an accurate model of the operation process of the motor train unit based on a data-driven ANFIS modeling method, transmitting the model to a generalized predictive controller, obtaining and outputting a control quantity u through specific calculation, and controlling the motor train unit to track a given inter-station operation mode curve (the operation mode curve is composed of an actual ATP speed limit curve and an optimal expected speed curve, wherein the optimal expected speed curve is obtained by screening and determining a large number of actual operation speed curves of the high-speed motor train unit based on operation indexes such as safety, right-point and energy conservation and in combination with the operation experience of excellent drivers). Real-time acquisition of predicted output speed y and desired speed y_rAnd analyzing the running speed error. When the high-speed motor train unit is influenced by uncertain factors such as unknown environment or motor train unit characteristic change and the like to track control, the absolute value | e of the feedback speed error is caused_rWhen the absolute value exceeds a set error threshold value x (which is selected by combining the allowable error range of the CTCS-3 train control system and the real-time control sampling period requirement) to return actual operation data, the model parameters are adjusted in real time by combining a Kalman filtering algorithm and a BP gradient descent method, and the generalized predictive controller of the high-speed motor train unit is adjusted based on the adjusted ANFIS model to adapt to the next high-speed motor train unitAnd the high-precision tracking control is performed, and the influence of characteristic change on the running of the motor train unit is eliminated.

2. Modeling the high-speed motor train unit in an off-line ANFIS mode in the operation process:

the stress condition of the high-speed motor train unit in the running process is shown in figure 2, a train is simplified into a single rigid mass point, all stress in the running process of the motor train unit is applied to the mass point for analysis and calculation, y in the figure is the running speed of the high-speed train and is obtained by a speed and distance measuring unit, u is unit control force (traction force/brake force) and is obtained by a driver operating a handle under the guidance of ATP vehicle-mounted equipment at present, and therefore the effects of traction, constant speed, coasting and braking are achieved, and r is_bIs unit basic resistance, r_b＝A_r+B_ry+C_ry². The stress condition of the high-speed motor train unit in the figure can be described by the following mathematical model.

In the formula, is the acceleration coefficient, A_r、B_r、C_rIs the coefficient of resistance, C_ry²Representing air resistance, as the speed of train operation increases, C_ry²The larger the proportion of the nonlinear optical fiber, the more obvious the nonlinear characteristic of the system is. The differential transformation of equation (1) can be described as the relation:

y(k)＝f{y(k-1),u(k-1)} (2)

in view of the fact that ANFIS integrates the adaptive learning characteristic of a neural network and the nonlinear modeling characteristic of a T-S fuzzy model, the model conclusion part of the ANFIS replaces fuzzy numbers in a general Mamdani fuzzy system with linear equations, so that the system can describe a complex nonlinear system with fewer rules. For the running process of the motor train unit described by the formula (2), the following fuzzy inference rule is adopted to describe

R_iExpressing the ith fuzzy inference rule; y (k-1), u (k-1) are input quantities, and y (k) is an output quantity;is the ith fuzzy set of input quantities;ξ for back-end parameters, n being the number of rules_iIs a constant term.

Suppose that m high-speed motor train unit operation process data points { X }₁,…,X_i,…,X_mIn which X is_i＝[v_i(k-1),u_i(k-1),v_i(k)]Data point X_iThe density index of (A) is defined as

In the formula,_athe radius of the effective neighborhood of the cluster center is set to be a positive number. Selecting the highest value of the density indexObtaining a first cluster center

Each data point X_iThe density index of (A) is corrected by the following formula

Wherein_bIs one is greater than_aPositive number of (c). Obviously, close to the first cluster center c₁The density indicator of the data points of (a) will be significantly reduced such that these points are less likely to be selected as the next cluster center. After the density index of each data point is corrected, the next clustering center c is selected₂And correcting all density indexes of the data points again, and repeating the process until the density indexes of the data points are corrected againGet the last cluster center c_nTherefore, the number of the clustering centers of the data in the running process of the high-speed motor train unit is n.

And obtaining n linear models by adopting minimum variance estimation on the n rule postresults. The n linear models were fused to obtain the following ANFIS model:

wherein,is the normalized value of all rule fitness.

3. High-speed motor train unit tracking operation real-time optimization control

Based on the established ANFIS model of the high-speed motor train unit, the corresponding generalized predictive controller is designed as follows:

the high-speed motor train unit operation process model (7) obtained in the above way can be described in the following form

In the formula,andis z^-1A polynomial of (1-z) (. DELTA. -)^-1。

Wherein the parametersAndobtained by a modeling process, which can be expressed as

Andis the model order of the controlled object.

To obtain the optimal control scalar, the performance indicator function can be designed as

Wherein, L, H, G andis the introduced parameter matrix of the loss-of-image equation,is a weighting coefficient matrix.

Minimize performance index (i.e.) Obtain the optimal control increment of

In order to eliminate the influence of the change of the running characteristics of the motor train unit caused by the unmodeled part, unknown environment and faults on the tracking control and feed back control errors, an ANFIS model online adjustment strategy is implemented:

step 1, designing a generalized predictive controller (as shown in the above) based on an ANFIS model of the high-speed motor train unit, and calculating a speed feedback error e at the time t_r(t)＝y(t)-y_r(t)。

And Step 2, executing the tracking operation control of the high-speed motor train unit.

Step 3, judging whether the time T reaches the total time T or not_tIf the process is finished, the process is ended, otherwise, the next time t +1 is entered, and the process is shifted to Step 4.

At the moment of Step 4, t +1, the feedback error e of the previous moment is calculated_r(t) comparing with a set error threshold χ, if | e_r(t) | is more than or equal to χ, the model mismatch is indicated, the model mismatch is not suitable for the control condition at that time, and the Step5 is required to carry out online adjustment on the model parameters; if the feedback error | e_r(t)|<And χ, the tracking precision is high, the model parameters are not needed to be optimized temporarily, and the Step 1 is directly returned.

And Step5, combining a Kalman filtering algorithm and a BP gradient descent method, and adjusting parameters of a front part and a back part of the ANFIS model on line (specifically shown as follows). And substituting the optimized model into the generalized predictive controller, and returning to Step 1.

On the basis of the established model (8), firstly, the following Kalman filtering algorithm is adopted to adjust the parameters of the back part

In the formula,a forgetting factor of 0 for a back-piece parameter obtained during modeling<λ_KF1A positive number close to 1 (in this context. lambda.) is generally chosen_KF＝0.9995)；P_KF(k)＝q_KFI∈R^2n×2n，q_KFIs a large positive number, usually taken as 10⁴～10¹⁰(in the invention q)_KF＝10⁶)。

Secondly, optimizing the front-part parameter c in real time by adopting a BP gradient descent method_ijAnd σ_ij。

Calculating an error index function of

Wherein the k-th data points y (k) and y_r(k) Respectively representing the actual output speed and the expected output speed at the moment t of the mismatch of the model transmitted by the control part, and so on. The front-part parameter optimization algorithm is as follows:

after the ANFIS model parameters are corrected, the ANFIS model parameters are substituted into the generalized predictive controller to be recalculated, corresponding control force is obtained, and real-time optimization control is implemented on the tracking operation of the high-speed motor train unit, and the specific flow is shown in FIG. 3.

In conclusion, aiming at the characteristics of complex running environment, frequent working condition change, more factors influencing running performance and the like of the high-speed motor train unit, an ANFIS model in the running process is established, and the tracking running real-time optimization control method of the high-speed motor train unit is provided. When the tracking performance of the motor train unit is deteriorated due to the characteristics or environmental changes of the motor train unit, an online adjustment strategy is started, the control performance is optimized in real time, and the safety and the punctuality of the motor train unit are improved.

The CRH380AL type high-speed motor train unit is selected as an experimental verification object in the implementation of the invention. Firstly, acquiring the whole-course running speed and control force data of the motor train unit in a section from West Jinan to Xuzhou east of the high-speed rail of Jinghu, selecting 2000 groups of effective data representing all working conditions of traction, coasting and braking, taking 1400 groups of data as modeling data samples on the whole average, and taking the remaining 600 groups of data as data for testing the model accuracy. First, model sample data is constructed according to 1400, and an ANFIS modeling method is adopted to obtain 4 optimal fuzzy rules (i.e. n is 4 in formula (3)), and the parameters of the offline ANFIS model are shown in table 1. To verify the validity of the model, the model was tested using the remaining 600 sets of operating data, and the error distribution curve of the model output was as shown in fig. 4. :

TABLE 1 ANFIS model parameters

The speed limit curve in fig. 4 is drawn according to the positioning speed measurement requirement of the CTCS-3 train control system. Observing the model validation process of FIG. 4, when the speed is less than 30km/h, the model outputs an error range: -0.5876-0.5234 km/h, when the speed is more than 30km/h, the error range is as follows: and the parameters are-2.1421-1.6899 km/h, the positioning speed measurement requirements of the CTCS-3 train control system are met, and the established ANFIS model is high in precision, strong in generalization capability and good in prediction effect.

Based on the established ANFIS model, the real-time optimization control strategy is designed by utilizing generalized predictive control and combining a Kalman filtering algorithm and a BP gradient descent method to implement real-time optimization control on the tracking operation of the high-speed motor train unit in the interval of the West station of Jinghai high-speed rail line, namely the Xuzhou east station. In the actual running process of the motor train unit, if the motor train unit characteristic changes due to an unknown environment, the tracking performance of the high-speed motor train unit is poor, so that an upper target curve is difficult to track, and a speed tracking error exceeds a set threshold value (a set threshold value χ is 2 km/h). Under the condition, the real-time optimization control strategy of the invention optimizes the running process model of the motor train unit in time, so that the high-speed motor train unit can quickly track the upper target curve again. When the motor train unit runs to the mileage of 500km and the mileage of 600km, uncertain interference factors are added, so that the running characteristics of the motor train unit are changed, the control simulation result is shown in fig. 5-8, and the optimization process curve of the real-time optimization process ANFIS model parameters is listed in fig. 9.

FIGS. 5-8 show that the speed changes suddenly and the target curve cannot be tracked under the condition that the motor train unit has changed characteristics. The ANFIS-based real-time optimization control strategy can optimize an operation model in real time according to the existing motor train unit data, and adjust the control force, so that the high-speed motor train unit can quickly correct the operation speed, and an upper target curve is tracked again with high precision. FIG. 9 shows that, after a fault, the method of the present invention applies to the context parameters of the ANFIS modelc_ij,σ_ijAnd (i is 1,2,3, 4; j is 1,2) carrying out real-time accurate adjustment until the optimal ANFIS model is obtained again, and optimizing the high-speed motor train unit operation model in real time. Table 2 lists the velocity and acceleration tracking errors of the inventive method after the unknown fault occurred.

TABLE 2 speed and acceleration tracking error

Feature change	Speed (kilometer/hour)	Acceleration (meter/square second)
			Maximum negative error	-1.6502	-0.4649
Maximum positive error	2.3763	0.7095
			Root mean square error	0.7793	0.0203

From table 2, it can be seen that, under the condition that the running characteristics of the motor train unit change, the maximum positive and negative tracking error and the root mean square error of the method are controlled within a certain range, the positioning and speed measuring requirements of the CTCS-3 train control system are met, and the effectiveness of the method is further quantitatively shown.

It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.

Claims (5)

1. A real-time optimization control method for a high-speed motor train unit is characterized in that the method comprises the steps of establishing an ANFIS model of the high-speed motor train unit by collecting actual operation data of the high-speed motor train unit, designing a corresponding prediction controller based on the ANFIS model, collecting predicted output speed and expected speed in real time, and analyzing an operation speed error; when the high-speed motor train unit is influenced by uncertain factors such as unknown environment or motor train unit characteristic change and the like on tracking control, the Kalman filtering algorithm and the BP gradient descent method are combined to adjust model parameters in real time so as to adapt to the following high-precision tracking control of the high-speed motor train unit, eliminate the influence of characteristic change on the operation of the motor train unit and ensure the operation performance of the high-speed motor train unit.

2. The real-time optimization control method for the high-speed motor train unit according to claim 1, wherein the offline ANFIS model of the high-speed motor train unit is as follows:

suppose that m high-speed motor train unit operation process data points { X }₁,…,X_i,…,X_mIn which X is_i＝[v_i(k-1),u_i(k-1),v_i(k)]Data point X_iThe density index of (A) is defined as

$D_i=\underset{j=1}{\overset{m}{\&Sigma;}}\exp \&lsqb;-\frac{\left|\right|X_i-X_j|{|}^2}{{({\mathrm{\&delta;}}_a/2)}^2}\&rsqb;,(i,j=1,2,...,m)$

In the formula,_athe radius of the effective neighborhood of the set clustering center is a positive number; selecting the highest value of the density indexObtaining a first cluster center

$c_1=X_i{|}_{{\mathrm{maxD}}_i}$

Each data point X_iThe density index of (A) is corrected by the following formula

$D_i=D_i-D_{c_1}\exp \&lsqb;-\frac{\left|\right|X_i-c_1|{|}^2}{{({\mathrm{\&delta;}}_b/2)}^2}\&rsqb;$

Wherein_bIs one is greater than_aA positive number of; obviously, close to the first cluster center c₁The density indicator of the data points of (a) will be significantly reduced, thus making it less likely that these points will be selected as the next cluster center; after the density index of each data point is corrected, the next clustering center c is selected₂Correcting all density indexes of the data points again, and repeating the processProcedure untilGet the last cluster center c_nTherefore, the number of the clustering centers of the data in the running process of the high-speed motor train unit is n;

obtaining n linear models by adopting minimum variance estimation on n rule postresults; the n linear models are fused to obtain the following offline ANFIS model:

$y\left(k\right)=\underset{i=1}{\overset{n}{\&Sigma;}}{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_i\&CenterDot;y_i\left(k\right)=\underset{i=1}{\overset{n}{\&Sigma;}}{\stackrel{\&OverBar;}{\mathrm{\&omega;}}}_i\&lsqb;{\mathrm{\&theta;}}_1^{i}y(k-1)+{\mathrm{\&theta;}}_2^{i}u(k-1)+{\mathrm{\&xi;}}_i\&rsqb;---\left(7\right)$

wherein,is the normalized value of all rule fitness.

3. The real-time optimal control method for the high-speed motor train unit according to claim 2, wherein the prediction controller based on the offline ANFIS model is as follows:

the running process model (7) of the high-speed motor train unit can be described in the following form

$\stackrel{\&OverBar;}{a}\left(z^{-1}\right)y\left(t\right)=\stackrel{\&OverBar;}{b}\left(z^{-1}\right)u(t-1)+\mathrm{\&xi;}\left(t\right)---\left(8\right)$

In the formula,andis z^-1The polynomial of (a) is determined,Δ＝1-z^-1；

$\stackrel{\&OverBar;}{a}\left(z^{-1}\right)=(1-z^{-1})a\left(z^{-1}\right)=1+{\stackrel{\&OverBar;}{a}}_1{z}^{-1}+...+{\stackrel{\&OverBar;}{a}}_{n_{\stackrel{\&OverBar;}{a}}+1}{z}^{-n_{\stackrel{\&OverBar;}{a}}},$

$\stackrel{\&OverBar;}{b}\left(z^{-1}\right)=(1-z^{-1})b\left(z^{-1}\right)={\stackrel{\&OverBar;}{b}}_0+{\stackrel{\&OverBar;}{b}}_1{z}^{-1}+...+{\stackrel{\&OverBar;}{b}}_{n_{\stackrel{\&OverBar;}{b}}}{z}^{-n_{\stackrel{\&OverBar;}{b}}},$

wherein the parametersAndobtained by a modeling process.

To obtain the optimal control scalar, the performance indicator function can be designed as

$\begin{array}{c}J={\&lsqb;L\mathrm{\&Delta;}U(t+j-1)+H\mathrm{\&Delta;}U(t-j)+GY\left(t\right)-\stackrel{\&OverBar;}{W}{Y}_r\left(t\right)\&rsqb;}^T\\ \stackrel{\&OverBar;}{\mathrm{\&mu;}}\&lsqb;L\mathrm{\&Delta;}U(t+j-1)+H\mathrm{\&Delta;}U(t-j)+GY\left(t\right)-\stackrel{\&OverBar;}{W}{Y}_r\left(t\right)\&rsqb;\\ +u^{T}ru\end{array}$

Wherein, L, H, G andis the introduced parameter matrix of the loss-of-image equation,is a weighting coefficient matrix;

minimize performance index (i.e.) Obtain the optimal control increment of

$\mathrm{\&Delta;}U(t+j-1)={(L^T\stackrel{\&OverBar;}{\mathrm{\&mu;}}L+R)}^{-1}{L}^T\stackrel{\&OverBar;}{\mathrm{\&mu;}}\&lsqb;\stackrel{\&OverBar;}{W}{Y}_r(t+j)-GY\left(t\right)-H\mathrm{\&Delta;}U(t-j)\&rsqb;.$

4. The real-time optimization control method of the high-speed motor train unit according to claim 3, wherein when the high-speed motor train unit is influenced by uncertain factors such as unknown environment or motor train unit characteristic change on tracking control, an online model adjusting strategy is started to adapt to the following high-precision tracking control of the high-speed motor train unit, the model parameters are adjusted in real time by combining a Kalman filtering algorithm and a BP gradient descent method through the real-time optimization control strategy, and the specific optimization steps can be represented as:

step 1, designing a generalized prediction controller based on an ANFIS model of the high-speed motor train unit, and calculating a speed feedback error e at the time t_r(t)＝y(t)-y_r(t)；

Step 2, executing the tracking operation control of the high-speed motor train unit;

step 3, judging whether the time T reaches the total time T or not_tIf the current time reaches the preset time, ending the process, otherwise entering the next time t +1, and turning to Step 4;

at the moment of Step 4, t +1, the feedback error e of the previous moment is calculated_r(t) and a set error threshold χBy comparison, if | e_r(t) | is more than or equal to χ, the model mismatch is indicated, the model mismatch is not suitable for the control condition at that time, and the Step5 is required to carry out online adjustment on the model parameters; if the feedback error | e_r(t)|<Chi indicates that the tracking precision is higher, the model parameters do not need to be optimized temporarily, and the model parameters directly return to Step 1;

and Step5, combining a Kalman filtering algorithm and a BP gradient descent method, adjusting parameters of a front part and a back part of the ANFIS model on line, substituting the optimized model into the generalized predictive controller, and returning to Step 1.

5. The real-time optimization control method for the high-speed motor train unit according to claim 4, wherein the Step5 is combined with a Kalman filtering algorithm and a BP gradient descent method, and the method for adjusting parameters of a front piece and a back piece of the ANFIS model on line comprises the following steps:

on the basis of the established model (8), firstly, the following Kalman filtering algorithm is adopted to adjust the parameters of the back part

In the formula,a forgetting factor of 0 for a back-piece parameter obtained during modeling<λ_KF1A positive number close to 1 (in this context. lambda.) is generally chosen_KF＝0.9995)；P_KF(k)＝q_KFI∈R^2n×2n，q_KFIs a large positive number, usually taken as 10⁴～10¹⁰Preferably q_KF＝10⁶；

Secondly, optimizing the front-part parameter c in real time by adopting a BP gradient descent method_ijAnd σ_ij；

Calculating an error index function of

$E=\frac{1}{2}{\left(y\right(k)-y_r(k\left)\right)}^2=\frac{1}{2}{\left(e_r\right(k\left)\right)}^2$

Wherein the k-th data points y (k) and y_r(k) Respectively representing the actual output speed and the expected output speed of the model transmitted by the control part at the moment of mismatch t, and so on; the front-part parameter optimization algorithm is as follows:

$c_{ij}(k+1)=c_{ij}\left(k\right)-{\mathrm{\&alpha;}}_c\frac{\&part;E}{\&part;c_{ij}}$

${\mathrm{\&sigma;}}_{ij}(k+1)={\mathrm{\&sigma;}}_r\left(k\right)-{\mathrm{\&alpha;}}_{\mathrm{\&sigma;}}\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{ij}}$

and after the ANFIS model parameters are corrected, substituting the corrected ANFIS model parameters into the generalized predictive controller for recalculation to obtain corresponding control force to implement real-time optimization control on the tracking operation of the high-speed motor train unit.