An energy-saving traction data-driven adaptive control method for urban rail transit trains

Abstract

This invention discloses a data-driven adaptive control method for energy-saving traction of urban rail transit trains. First, based on the principle of a kernel limit learning machine autoencoder, a measurement data feature model is constructed by scaling the urban rail transit train operation data. Then, feature vectors of offline optimization results and online measurement data are obtained. Based on the measurement data feature model and model-free adaptive control, a recursive adaptive control law for online following of the train's energy-saving curve is established. Next, the adaptive control law for online following of the energy-saving traction curve based on data feature control is obtained. Finally, the process control variables for online following of the energy-saving traction curve are obtained. This invention introduces kernel limit learning machine and model-free adaptive control into the online following process of the energy-saving traction curve, using data feature control and transfer as the main line, and obtains the adaptive control law for online following of the energy-saving traction curve based on deep data-driven methods, thereby improving the overall stability, adaptability, and control accuracy of online following of the energy-saving traction curve.

Description

Technical Field

[0001] This invention relates to the field of urban rail transit train control technology, and particularly to an energy-saving traction data-driven adaptive control method for urban rail transit trains. Background Technology

[0002] Train energy conservation comprises two parts: offline optimization and online following. Offline optimization of the train's energy-saving traction curve is the foundation of energy-saving train operation optimization, while online following control is crucial for executing energy-saving traction parameters. Online following of the train traction curve mainly involves online following control of the recommended speed. Due to the combined effects of large inertia, nonlinearity, uncertainty, and randomness, online following control of the recommended speed of the train traction curve is difficult to achieve directly through mechanism modeling and traditional control methods.

[0003] Due to the uncertain evolution of the train operating environment and the complex coupling of factors affecting train operation control, optimization control methods relying on the mathematical model or nominal model of the controlled object are difficult to adapt to the online following process of the traction curve, where the control structure is constantly evolving and the control parameters are uncertain. Data-driven control methods can establish a recursive control law for the controlled system based on the input and output data of the controlled system, through the mapping relationship between control quantities and measurements, thus achieving adaptive control of the controlled object. The offline energy-saving optimization results of trains are mostly parameter combination sequences; relying solely on the measured value of train speed cannot represent the data characteristics of offline optimization. Direct feedback control through the measured variables of the traction control system can easily cause the operating state of the control target to deviate from the offline optimized state data of the energy-saving traction curve, easily resulting in the operating state data of the control process being ahead or behind, ultimately leading to a serious reduction in the online following performance of the energy-saving traction curve.

[0004] Due to the high computational complexity of energy-saving optimization in urban rail transit systems, and constrained by factors such as train operation safety regulations, uncertainties in operating lines, and the efficiency of energy-saving optimization iterations, energy-saving optimization in urban rail transit systems is actually achieved through a combination of offline optimization and online tracking control. The offline optimization results of train energy-saving parameters exist as a sequence of parameter combinations. Current technologies for implementing online tracking control of train energy-saving traction curves rely on direct sensor measurements, directly feeding back control through the measured variables of the traction control system. Feedback control based on direct measurement data depends on the accuracy of sensor measurements. Due to the complexity of factors influencing train operation, sensor measurements inevitably suffer from some degree of inaccuracy.

[0005] The shared relationships and coupling constraints among various components and input / output variables during train operation result in extremely complex relationships between variables, making it difficult to describe the traction curve online following control system using intuitive mathematical expressions. A precise mathematical model is the ideal method for achieving optimal train operation control. However, actual train traction is a complex motion system with interdependent and mutually restrictive factors such as nonlinearity, uncertainty, time-varying nature, and incompleteness, making it difficult to obtain a precise mathematical model. Due to uncertainties such as wear of transmission components, changes in traction motor performance, passenger capacity variations, and changes in the operating environment, the uncertainty of various parameters and the variability of the controlled model accompany every stage of the train integrated automatic control system's perception, decision-making, and control processes. Given the complex influencing factors and their intricate coupling relationships during train operation, conventional algorithms based on classical control theory struggle to achieve satisfactory automatic control results.

[0006] Online tracking of the train's energy-saving optimization curve can be directly achieved through feedback control of measured variables in the traction control system. However, since most train energy-saving optimization curves are obtained based on offline optimization, and due to the complexity of factors affecting train energy consumption, online tracking of the energy-saving traction curve relying solely on speed measurements may not effectively unify the parameter combinations obtained from offline optimization. This can easily lead to the actual tracking state of energy-saving parameters deviating from the offline optimization results, ultimately reducing the overall effectiveness of the train's energy-saving optimization control process. Furthermore, the complexity of factors affecting train energy consumption means that online tracking of the energy-saving traction curve relying solely on a single measurement value may not effectively unify the parameter combinations obtained from offline optimization. This can easily lead to the actual tracking state of energy-saving parameters deviating from the offline optimization results, and even cause tracking control instability, with potentially incalculable consequences.

[0007] Currently, the mainstream control methods for online traction curve following mainly include PID control, fuzzy control, model reference control, and predictive control. Most of these methods rely on direct feedback control based on sensor measurement data, neglecting the constraints of data characteristics and exhibiting various limitations. PID control-based online traction curve following is simple in structure and easy to implement, but it can only achieve target control and cannot adapt to changes in the operating environment. Fuzzy control-based online traction curve following is susceptible to the influence of fuzzy rule formulation, resulting in low control accuracy, inability of fuzzy rule quantization parameters to adapt to environmental changes, and an inability to balance precision and speed control. Model reference control-based online traction curve following relies on a train operation process control model; however, the numerous and complex coupling factors affecting train operation control lead to many limitations. Predictive control-based online traction curve following depends on process models and mathematical expressions, facing modeling difficulties and model mismatch issues in the control process, resulting in limitations in model construction for practical adaptive control. Summary of the Invention

[0008] The purpose of this invention is to provide an energy-saving traction data-driven adaptive control method for urban rail transit trains, which can solve problems such as difficulty in obtaining accurate mathematical models for online tracking of energy-saving traction curves, inability to adapt to changes in the operating environment, large steady-state errors and low tracking performance, and inability to balance low control accuracy and fast control.

[0009] To achieve the above objectives, the present invention provides an energy-saving traction data-driven adaptive control method for urban rail transit trains, comprising the following steps:

[0010] Step S1: Collect online sensor data on the operating status of urban rail transit trains. Obtain historical load statistics and traction energy-saving offline optimization results.

[0011] Step S2: The urban rail transit train operation status data and historical transport statistics data obtained in step S1 are used as the basic dataset of the kernel limit learning machine. The energy-saving traction curve process status and data are mapped to a high-dimensional space. The data feature vector of the control target is obtained from the high-dimensional space. Based on the principle of kernel limit learning machine autoencoder, the urban rail transit train operation data is scaled to construct a measurement data feature model.

[0012] Step S3: Based on the urban rail transit train operation status data measured online by the sensors obtained in Step S1, the traction energy saving offline optimization results, and the measurement data feature model obtained in Step S2, obtain the offline optimization result data feature vector and the online measurement data feature vector.

[0013] Step S4: Based on the measurement data feature model obtained in step S2, establish an online following recursive adaptive control law for the energy-saving curve of urban rail transit trains based on the measurement data feature model and model-free adaptive control.

[0014] Step S5: Taking the control and transfer of the data feature vector obtained in step S3 as the main line, the model-free adaptive control principle is applied to the data feature following control process of the energy-saving traction curve to obtain the online following adaptive control law of the energy-saving traction curve based on data feature control.

[0015] Step S6: Based on the adaptive control law for online following of the energy-saving traction curve obtained in step S5 and the data feature vector obtained in step S3, obtain the process control quantity for online following of the energy-saving traction curve.

[0016] Step S7: Implement amplitude limiting feedback output on the process control quantity obtained in step S6, and suppress and iteratively correct the process control quantity based on the amplitude limiting feedback re-output strategy to meet the actual equipment's tolerance.

[0017] Furthermore, in step S1, the urban rail transit train operation status data includes: train operation acceleration, train operation speed, actual train operation distance measurement, train traction motor input power, train traction motor electrical phase angle and train traction torque; historical load statistics data includes: historical load data and historical running inertia data; traction energy saving offline optimization results are operating condition transition point parameters, including train operation distance, train acceleration and train speed.

[0018] Furthermore, in step S2, the method for constructing the measurement data feature model is as follows:

[0019] The urban rail transit train operation status data and historical transport statistics obtained in step S1 are used as the basic dataset for the kernel extreme learning machine. The raw data vector of train operation data is denoted as The vector obtained after scaling the historical train operation data vector is denoted as... ;

[0020] The principle of kernel limit learning machine autoencoder is applied to the modeling process of measurement data feature model, and the original data vector of train operation data is transformed. The encoding function serves as the input to the kernel extreme learning machine. vector Encode as a vector Decoding function vector Decode into vector ,make Reproduction This enables the scaling transformation of train data and the extraction of data features; the mathematical description of the process of scaling transformation of train operation data and extraction of data features is as follows.

[0021] ,

[0022] ,

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] ,

[0028]

[0029] Represents the feature vector of the measurement data; For the hidden layer The connection weights of each neuron For the corresponding number Activation function of a neuron For the weight vector, For bias terms; Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix; The kernel function represents the machine learning method.

[0030] Furthermore, in step S4, the method for establishing an online following recursive adaptive control law for the energy-saving curve of urban rail transit trains based on the characteristic model of measurement data and model-free adaptive control is as follows:

[0031] The principle of model-free adaptive control is applied to the online following recursive control law construction process; based on the principle of model-free adaptive control, the mathematical description of the online following control process of the energy-saving traction curve of urban rail transit trains is as follows.

[0032] ,

[0033] ,

[0034] ,

[0035] and These respectively indicate the train's position at the... The target feature vector and process control vector at each moment; and Two positive integers, representing the system order; It is a non-linear function; It is a constraint control matrix that is highly time-varying and related to the input and output data of the train's online tracking process for energy-saving curves; it is also known as a pseudo-partial derivative matrix. Indicates the train is at The increment of the process control parameter sequence at each time step; based on the principle of model-free adaptive control, this invention establishes the following target control optimization criterion:

[0036] ,

[0037] Indicates about the first Time process control vector The optimization objective function; Indicates the first Expected data characteristics at any given time; and These are the weights of the optimization objective; Indicates the acquisition and control cycle; and Substitute into the optimization criterion function ,right Taking the partial derivative and setting it equal to 0, the process control vector is... The recurrence relation is as follows:

[0038] ,

[0039] This represents the iteration step size factor of the control process; appropriately adjusting its value can improve the response speed of the control system. In the recursive calculation formula for process control variables, The value of is related to the overall control effect and is estimated online based on the input and output data of the control system; in order to estimate the matrix Based on the principle of model-free adaptive control, this invention focuses on minimizing the difference in train speed following deviation within adjacent acquisition and control cycles. It uses the estimated value of the pseudo-partial derivative matrix to approximate its true value as a constraint, constructing a model regarding... The optimization criterion function has the following specific form:

[0040] ,

[0041] It is about the pseudo-partial derivative matrix The optimization objective function; It optimizes the objective weights; it adjusts the objective function. Find the partial derivatives and set them equal to 0, then the estimated matrix of the pseudopartial derivative matrix is ​​obtained. The recursive formula is shown below:

[0042] ,

[0043] It is the iteration step size control factor; express The change in the actual train speed at any given time; combining the identification process of the pseudo-partial derivative matrix with the process control quantity, and introducing a constant. For stable process control, the complete recursive control law for online adaptive control of energy-saving curve following of urban rail trains based on data characteristics is as follows:

[0044] ,

[0045] .

[0046] Furthermore, in step S7, the process control quantity obtained in step S6 is subjected to amplitude limiting feedback output. The method for suppressing and iteratively correcting the process control quantity based on the amplitude limiting feedback re-output strategy is as follows:

[0047] When the calculation result of the recursive adaptive control law exceeds the equipment's tolerance limit, the limiting output information is fed back to the recursive calculation formula of the process control quantity, and then the recursive control law is used to calculate the output, ensuring that the process control quantity output by the control algorithm is limited to the range that the equipment can tolerate, and the limiting output data is integrated into the process control law so that the limiting output information of the previous moment can be perceived when the next control cycle performs iterative calculation.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] (1) This invention replaces direct measurement feature control with energy-saving traction parameter data feature control and transfer, avoiding the problem that online direct measurement feedback control cannot take into account offline optimization of each data component, and improving the overall effect of online follow-up control of energy-saving parameters;

[0050] (2) The present invention is based on the deep data-driven acquisition of the energy-saving traction curve online following adaptive control law, which effectively solves the problem that it is difficult to realize the energy-saving traction curve online following control based on mechanism modeling;

[0051] (3) This invention does not rely on the precise mathematical model of the controlled object, thus avoiding the problem of control failure caused by model reference control and predictive control techniques due to modeling difficulties and model mismatch;

[0052] (4) This invention combines energy-saving parameter data characteristics with model-free adaptive control, avoiding problems such as the inability of fuzzy control fuzzy rule parameters to learn changes in the train operating environment and the difficulty of PID control to adapt to the continuous evolution of the train operating environment.

[0053] (5) The present invention introduces a limiting feedback and re-output strategy, which limits the output and feeds back the limiting information to the control law at the same time. This not only effectively solves the problem that the actual equipment cannot withstand the process control quantity, but also improves the stability of the adaptive control law. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart of the energy-saving traction data-driven adaptive control method for urban rail transit trains according to the present invention;

[0056] Figure 2 Network diagram for building a feature model of measurement data;

[0057] Figure 3 A structural diagram for building a feature model of measurement data;

[0058] Figure 4 This is a model-free adaptive control block diagram based on a feature model of measurement data;

[0059] Figure 5 The process of limiting feedback and re-output of process control quantities;

[0060] Figure 6 The implementation process of online following control for train energy-saving traction curves;

[0061] Figure 7 This is a diagram illustrating the speed following control effect of the present invention. Detailed Implementation

[0062] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0063] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0064] Example 1

[0065] Please see Figure 1 An energy-saving traction data-driven adaptive control method for urban rail transit trains includes the following steps:

[0066] Step S1: Collect online sensor data on the operating status of urban rail transit trains. Obtain historical load statistics and traction energy-saving offline optimization results.

[0067] The urban rail transit train operation status data includes: train acceleration, train speed, actual train running distance measurement, train traction motor input power, train traction motor electrical phase angle and train traction torque; historical load statistics data includes: historical load data and historical running inertia data; traction energy saving offline optimization results are operating condition transition point parameters, including train running distance, train acceleration and train speed;

[0068] Collecting urban rail transit train operation status data and acquiring historical load statistics are crucial for effective train control. Train acceleration is a direct control variable that changes train speed and a key state parameter in offline optimization. Whether in offline optimization or online tracking, train acceleration is a vital component of the energy-saving traction curve parameter feature space. Train travel distance is not only a key parameter in offline optimization but also a major factor influencing control state and limiting control parameter values ​​during online tracking of the traction curve. Although measured train speed values ​​may match recommended values, different travel distances result in different data characteristics. Traction force, traction torque, and motor drive current are dominant variables affecting acceleration during actual train operation; integrating their historical data characteristics into the measurement data feature model helps improve the rational selection of process control variables. Train inertia influences the response speed of the online tracking control process for the recommended speed of the energy-saving traction curve. Train inertia primarily depends on the train's weight and structure, but actual load also has some impact. Integrating quantified historical load data helps the measurement data feature model calculate data characteristics that better reflect the actual transport environment. For the same train on the same line, the magnitude of train acceleration constrains the trend of train speed change. Traffic condition information from operating routes has a feedback effect on process control quantities. The feedback information from operating routes is quantified and incorporated into the measurement data feature model to improve the ability of process control laws to adapt to changes in the operating route environment.

[0069] Step S2: Using the urban rail transit train operation status data and historical transport statistics data obtained in Step S1 as the basic dataset for the kernel extreme learning machine, the energy-saving traction curve process state and data are mapped to a high-dimensional space. The data feature vector of the control target is obtained from the high-dimensional space. Based on the kernel extreme learning machine autoencoder principle, a scaling transformation is performed on the urban rail transit train operation data to construct a measurement data feature model. The network diagram of the measurement data feature model is shown below. Figure 2 As shown, the structure diagram of the measurement data feature model is as follows: Figure 3 As shown;

[0070] The method for constructing the feature model of measurement data is as follows:

[0071] The urban rail transit train operation status data and historical transport statistics data obtained in step S1 are used as the basic dataset for the kernel extreme learning machine. The raw data vector of train operation data is denoted as The vector obtained after scaling the historical train operation data vector is denoted as... ;

[0072] The principle of kernel limit learning machine autoencoder is applied to the modeling process of measurement data feature model, and the original data vector of train operation data is transformed. The encoding function serves as the input to the kernel extreme learning machine. vector Encode as a vector Decoding function vector Decode into vector ,make Reproduction This enables the scaling transformation of train data and the extraction of data features; the mathematical description of the process of scaling transformation of train operation data and extraction of data features is as follows.

[0073] ,

[0074] ,

[0075] ,

[0076] ,

[0077] ,

[0078] ,

[0079] ,

[0080] ,

[0081] Represents the feature vector of the measurement data; For the hidden layer The connection weights of each neuron For the corresponding number Activation function of a neuron For the weight vector, For bias terms; Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix; The kernel function represents the machine learning method.

[0082] Step S3: Based on the urban rail transit train operation status data measured online by the sensors obtained in Step S1, the traction energy saving offline optimization results, and the measurement data feature model obtained in Step S2, obtain the offline optimization result data feature vector and the online measurement data feature vector.

[0083] Among them, the traction energy-saving offline optimization result refers to the parameters of the operating condition transition point, including the train running distance, train acceleration, and train speed at the corresponding point; the data feature vector is a feature vector containing information such as train running distance, train acceleration, train speed, historical traction power, historical input voltage, line information, line curves, slopes, traction substations, and historical load capacity. It is the feature information representing the original data obtained after comprehensive processing of train running distance, train acceleration, train speed, line information, etc. Because it is a vector, it is called a feature vector.

[0084] Step S4: Based on the measurement data feature model obtained in Step S2, establish an online recursive adaptive control law for the energy-saving curve of urban rail transit trains based on the measurement data feature model and model-free adaptive control; The model-free adaptive control block diagram based on the measurement data feature model is shown below. Figure 4 As shown;

[0085] The method for establishing an online recursive adaptive control law for energy-saving curves of urban rail transit trains based on measurement data feature models and model-free adaptive control is as follows:

[0086] The principle of model-free adaptive control is applied to the online following recursive control law construction process; based on the principle of model-free adaptive control, the mathematical description of the online following control process of the energy-saving traction curve of urban rail transit trains is as follows.

[0087] ,

[0088] ,

[0089]

[0090] and These respectively indicate the train's position at the... The target feature vector and process control vector at each moment; and Two positive integers, representing the system order; It is a non-linear function; It is a constraint control matrix that is highly time-varying and related to the input and output data of the train's online tracking process for energy-saving curves; it is also known as a pseudo-partial derivative matrix. Indicates the train is at The increment of the process control parameter sequence at each time step; based on the principle of model-free adaptive control, this invention establishes the following target control optimization criterion:

[0091] ,

[0092] Indicates about the first Time process control vector The optimization objective function; Indicates the first Expected data characteristics at any given time; and These are the weights of the optimization objective; Indicates the acquisition and control cycle; and Substitute into the optimization criterion function ,right Taking the partial derivative and setting it equal to 0, the process control vector is... The recurrence relation is as follows:

[0093]

[0094] This represents the iteration step size factor of the control process; appropriately adjusting its value can improve the response speed of the control system. In the recursive calculation formula for process control variables, The value of is related to the overall control effect and is estimated online based on the input and output data of the control system; in order to estimate the matrix Based on the principle of model-free adaptive control, this invention focuses on minimizing the difference in train speed following deviation within adjacent acquisition and control cycles. It uses the estimated value of the pseudo-partial derivative matrix to approximate its true value as a constraint, constructing a model regarding... The optimization criterion function has the following specific form:

[0095] ,

[0096] It is about the pseudo-partial derivative matrix The optimization objective function; It optimizes the objective weights; it adjusts the objective function. Find the partial derivatives and set them equal to 0, then the estimated matrix of the pseudopartial derivative matrix is ​​obtained. The recursive formula is shown below:

[0097] ,

[0098] It is the iteration step size control factor; express The change in the actual train speed at any given time; combining the identification process of the pseudo-partial derivative matrix with the process control quantity, and introducing a constant. For stable process control, the complete recursive control law for online adaptive control of energy-saving curve following of urban rail trains based on data characteristics is as follows:

[0099] ,

[0100] .

[0101] Step S5: Taking the control and transfer of the data feature vector obtained in step S3 as the main line, the model-free adaptive control principle is applied to the data feature following control process of the energy-saving traction curve to obtain the online following adaptive control law of the energy-saving traction curve based on data feature control.

[0102] Step S6: Based on the adaptive control law for online following of the energy-saving traction curve obtained in step S5 and the data feature vector obtained in step S3, obtain the process control quantity for online following of the energy-saving traction curve.

[0103] Step S7: Implement amplitude limiting feedback output on the process control quantity obtained in step S6, and suppress and iteratively correct the process control quantity based on the amplitude limiting feedback re-output strategy to meet the actual equipment's tolerance.

[0104] Please see Figure 5 The method for suppressing and iteratively correcting the process control quantity based on the amplitude-limiting feedback output strategy is as follows:

[0105] When the calculation result of the recursive adaptive control law exceeds the equipment's tolerance limit, the limiting output information is fed back to the recursive calculation formula of the process control quantity, and then the recursive control law is used to calculate the output, ensuring that the process control quantity output by the control algorithm is limited to the range that the equipment can tolerate, and the limiting output data is integrated into the process control law so that the limiting output information of the previous moment can be perceived when the next control cycle performs iterative calculation.

[0106] The final online following control implementation process for the train's energy-saving traction curve is as follows: Figure 6 As shown, the speed following control effect achieved by this invention is as follows: Figure 7 As shown.

[0107] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A data-driven adaptive control method for energy-saving traction of urban rail transit trains, characterized in that, Includes the following steps: Step S1: Collect urban rail transit train operation status data measured online by sensors, obtain historical load statistics data, and obtain traction energy-saving offline optimization results; Step S2: The urban rail transit train operation status data and historical transport statistics data obtained in step S1 are used as the basic dataset of the kernel limit learning machine. The energy-saving traction curve process status and data are mapped to a high-dimensional space. The data feature vector of the control target is obtained from the high-dimensional space. Based on the principle of kernel limit learning machine autoencoder, the urban rail transit train operation data is scaled to construct a measurement data feature model. Step S3: Based on the urban rail transit train operation status data measured online by the sensors obtained in Step S1, the traction energy saving offline optimization results, and the measurement data feature model obtained in Step S2, obtain the offline optimization result data feature vector and the online measurement data feature vector. Step S4: Based on the measurement data feature model obtained in step S2, establish an online following recursive adaptive control law for the energy-saving curve of urban rail transit trains based on the measurement data feature model and model-free adaptive control. Step S5: Taking the control and transfer of the data feature vector obtained in step S3 as the main line, the model-free adaptive control principle is applied to the data feature following control process of the energy-saving traction curve to obtain the online following adaptive control law of the energy-saving traction curve based on data feature control. Step S6: Based on the adaptive control law for online following of the energy-saving traction curve obtained in step S5 and the data feature vector obtained in step S3, obtain the process control quantity for online following of the energy-saving traction curve. Step S7: Implement amplitude limiting feedback output on the process control quantity obtained in step S6, and suppress and iteratively correct the process control quantity based on the amplitude limiting feedback re-output strategy to meet the actual equipment's tolerance. In step S4, the method for establishing an online following recursive adaptive control law for the energy-saving curve of urban rail transit trains based on the characteristic model of measurement data and model-free adaptive control is as follows: The principle of model-free adaptive control is applied to the online following recursive control law construction process. Based on the principle of model-free adaptive control, the mathematical description of the online following control process of the energy-saving traction curve of urban rail transit trains is as follows: ， ， ， These respectively indicate the train's position at the... The target feature vector and process control vector at each moment; and Two positive integers, representing the system order; It is a non-linear function; It is a constraint control matrix that is highly time-varying and related to the input and output data of the train's online tracking process for energy-saving curves; it is also known as a pseudo-partial derivative matrix. Indicates the train is at The increment of the process control parameter sequence at each time step; based on the principle of model-free adaptive control, the following target control optimization criterion is established: ， Indicates about the first Time process control vector The optimization objective function; Indicates the first Expected data characteristics at any given time; and These are the weights of the optimization objective; Indicates the acquisition and control cycle; Substitute into the optimization criterion function ,right Taking the partial derivative and setting it equal to 0, the process control vector is... The recurrence relation is as follows: ， This represents the iteration step size factor of the control process; appropriately adjusting its value can improve the response speed of the control system. In the recursive calculation formula for process control variables, The value of is related to the overall control effect and is estimated online based on the input and output data of the control system; in order to estimate the matrix Based on the principle of model-free adaptive control, this paper focuses on minimizing the difference in train speed following deviation within adjacent acquisition and control cycles. It uses the estimated value of the pseudo-partial derivative matrix to approximate its true value as a constraint, and constructs a model regarding... The optimization criterion function has the following specific form: ， It is about the pseudo-partial derivative matrix The optimization objective function; It is to optimize the target weight; Optimize the objective function pair Find the partial derivatives and set them equal to 0, then the estimated matrix of the pseudopartial derivative matrix is ​​obtained. The recursive formula is shown below: ， It is the iteration step size control factor; express The change in the actual train speed at any given time; combining the identification process of the pseudo-partial derivative matrix with the process control quantity, and introducing a constant. For stable process control, the complete recursive control law for online adaptive control of energy-saving curve following of urban rail trains based on data characteristics is as follows: ， 。 2. The energy-saving traction data-driven adaptive control method for urban rail transit trains according to claim 1, characterized in that, In step S1, the urban rail transit train operation status data includes: train running acceleration, train running speed, actual train running distance measurement, train traction motor input power, train traction motor electrical phase angle and train traction torque; historical load statistics data includes: historical load data and historical running inertia data; traction energy saving offline optimization results are operating condition transition point parameters, including train running distance, train acceleration and train speed.

3. The energy-saving traction data-driven adaptive control method for urban rail transit trains according to claim 1, characterized in that, In step S2, the method for constructing the feature model of the measurement data is as follows: The urban rail transit train operation status data and historical transport statistics obtained in step S1 are used as the basic dataset for the kernel extreme learning machine. The raw data vector of train operation data is denoted as The vector obtained after scaling the historical train operation data vector is denoted as... ; The principle of kernel limit learning machine autoencoder is applied to the modeling process of measurement data feature model, and the original data vector of train operation data is transformed. The encoding function serves as the input to the kernel extreme learning machine. vector Encode as a vector Decoding function vector Decode into vector ,make Reproduction This enables the scaling transformation and feature extraction of train operation data; the mathematical description of the scaling transformation and feature extraction process is as follows: ， ， ， ， ， ， ， ， Represents the feature vector of the measurement data; For the hidden layer The connection weights of each neuron For the corresponding number Activation function of a neuron For the weight vector, For bias terms; Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix; The kernel function represents the machine learning method.

4. The energy-saving traction data-driven adaptive control method for urban rail transit trains according to claim 1, characterized in that, In step S7, the process control quantity obtained in step S6 is subjected to amplitude limiting feedback output. The method for suppressing and iteratively correcting the process control quantity based on the amplitude limiting feedback re-output strategy is as follows: When the calculation result of the recursive adaptive control law exceeds the equipment's tolerance limit, the information of the limiting output is fed back to the recursive calculation formula of the process control quantity, and then the output is calculated using the recursive control law to ensure that the process control quantity output by the control algorithm is limited to the range that the equipment can tolerate. Furthermore, the limiting output data is integrated into the process control law so that the limiting output information of the previous moment can be perceived when iterating in the next control cycle.

Images