Adaptive control method and system based on train operating condition identification

Abstract

Disclosed in the present invention are an adaptive control method and system based on train operating condition identification. The method comprises: first, acquiring operating state data of a train under different traction conditions, and calculating traction and braking performance data; secondly, forming an adhesion condition feature-related dataset; then, on the basis of the adhesion condition feature-related dataset, classifying adhesion conditions of the train into Nz levels, wherein Nz is a natural number; and finally, on the basis of number Nz of adhesion condition levels of the train and the current operating state data, identifying the current adhesion condition level Nd of the train, wherein Nd is a natural number and Nd≤Nz, establishing a mapping relationship between an adhesion coefficient under the current adhesion condition level and the maximum train braking deceleration, and generating an optimal train protection curve. In the method, adhesion conditions are clustered and classified using a clustering algorithm, allowing for more accurate analysis of the adhesion conditions; and an optimal train protection curve is adaptively generated by means of adhesion condition levels, thereby improving the reliability and safety of train operation.

Description

An adaptive control method and system based on train operating condition identification

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese Patent Application No. CN202411028745.9, filed on July 30, 2024, entitled "An Adaptive Control Method and System Based on Train Operation Condition Identification", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention belongs to the field of rail transit technology, and specifically relates to an adaptive control method and system based on train operation condition identification. Background Technology

[0004] Trains operate in a variety of harsh and complex environments, including extreme cold, rain, snow, sandstorms, and high humidity. Under these complex operating conditions, the adhesion coefficient between the train's rails and wheels changes significantly, directly affecting the braking process and consequently, the train's operational safety. Therefore, accurate and reliable identification of adhesion conditions is crucial for the safe and efficient tracking operation of high-speed trains. The adhesion operating point (i.e., the real-time adhesion coefficient) is the primary driver of these changes. The optimal adhesion condition lies at the boundary between the stable and unstable regions of adhesion characteristics and varies with wheel-rail operating conditions. Accurate online identification of adhesion conditions is a key technology for automatic driving of high-speed trains. The setting of optimized control inputs for automatic driving of high-speed trains relies on accurate and reliable identification of adhesion conditions. During multi-car tracking operations of high-speed trains, the safe tracking interval is closely related to the wheel-rail adhesion condition. Accurate and reliable identification of adhesion conditions is essential for ensuring the safe and efficient close tracking operation of high-speed trains.

[0005] Furthermore, existing protection curve generation technologies (such as the European standard model Subset-026) primarily calculate protection curves by manually setting train deceleration values. These deceleration values ​​mainly consider five factors: train speed, basic resistance deceleration, deceleration mode and level, track wet / dry condition, and wind resistance. Based on the values ​​of these factors, the train braking force is statically selected using a lookup table method, generally choosing braking force data under the most unfavorable condition of wet rail as the basis for protection curve calculation. The actual train acceleration is further calculated by incorporating the influence of gradient, typically ignoring uphill sections when considering the most unfavorable conditions. This process usually fails to dynamically adjust the braking force and dynamically generate the protection curve based on changes in adhesion conditions. When the train output braking force is large, adverse operating conditions such as wheel lock-up can easily occur, leading to insufficient protection and affecting train operation safety. On the other hand, safety-oriented train protection typically sets large safety margins, such as adding reduction coefficients to the train deceleration selection for calculation. This can easily result in over-protection in the obtained protection curve, limiting train operating efficiency.

[0006] Therefore, how to dynamically generate train protection curves based on adhesion conditions to ensure train operation safety is an urgent problem to be solved in the fields of rail transit technology, train control system technology, and adhesion condition simulation technology.

[0007] Summary of the Invention

[0008] To address the aforementioned problems, this invention provides an adaptive control method and system based on train operating condition identification. This method and system exhibit strong adaptability, high safety, and high reliability.

[0009] The purpose of this invention is to provide an adaptive control method based on train operating condition identification, comprising:

[0010] A multi-sensor fusion adhesion feature state self-calibration observer is used to acquire train operating status data under different traction conditions, and traction and braking performance data are calculated.

[0011] By combining traction and braking performance data, track information data, and operation control information, a multi-sensor fusion adhesive condition feature-related dataset is formed.

[0012] Based on the adhesion condition feature-related dataset, the train adhesion condition is divided into N z There are several levels, of which N z It is a natural number;

[0013] Based on the number N of train adhesion condition levels z Based on the current operating status data, identify the current adhesion condition level N of the train. d N d N is a natural number, and d≤N z ;

[0014] Enter the current adhesion level of the train, N. d Establish the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level;

[0015] Based on the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level, a train protection curve is generated.

[0016] Furthermore, the calculated traction and braking performance data includes inputting the train's operating state data under different traction conditions into the corresponding train dynamics model. These different traction conditions include starting, acceleration, cruising, coasting, and braking phases. The train dynamics model under these different traction conditions satisfies the following:

[0017]

[0018] Where Y(t), X(t-1), and Z(t) are the model's output at time t, its internal state at time t-1, and its input at time t, respectively, and f i (X(t-1), Z(t)) represent the dynamic characteristics of the i-th traction condition at time t fitted using a deep learning algorithm, w i (t) represents the fitting error of the i-th traction condition at time t, where i≤5 and is an integer.

[0019] Furthermore, a multi-sensor fusion-based adhesion feature state self-calibration observer is used to acquire train operating status data under different traction conditions, including:

[0020] The train speed was measured using wheel axle speed sensors and Doppler radar speed sensors, respectively.

[0021] Train acceleration is obtained using an accelerometer sensor;

[0022] A humidity sensor and a temperature sensor are used to obtain the temperature and humidity in the environment, respectively.

[0023] Furthermore, the acquisition of train operating status data under different traction conditions using a multi-sensor fusion-based adhesion feature state self-calibration observer also includes:

[0024] Based on the measured train operating speed, speed test dataset V was obtained. r,j and V d，j , where V r，j V is the dataset corresponding to the train running speed measured by the wheel axle speed sensor. d,j This is a dataset corresponding to train speeds measured by Doppler radar speed sensors.

[0025] Based on dataset V r,j and V d,j The speed measurement deviation dataset E was calculated. r,j and train speed measurement dataset V c，j It can be described as follows:

[0026] V c，j =·r1*g(V r，j )+·r2*g(V d，j (1.2)

[0027] E r，j =·h(V r，j -V d，j (1.3)

[0028] In the formula, r1 and r2 are the self-adjusting weights for velocity fusion, and g(·) and h(·) represent the fusion strategy functions.

[0029] Furthermore, the route information data includes route longitudinal profile data;

[0030] The operational control information includes train control commands and temporary speed limits.

[0031] Furthermore, based on the relevant dataset of adhesion condition characteristics, the train adhesion condition is divided into N z The levels include using a pseudo-nearest neighbor algorithm based on fuzzy clustering to divide the train adhesion condition into N levels. z The levels specifically include:

[0032] Based on the traction and braking performance data output from the train dynamics model, an R is estimated. m value;

[0033] R was randomly selected from the dataset related to adhesion condition features. m One data point is used as the first centroid;

[0034] The Euclidean distance calculation method is used to calculate the distance from each data point in the adhesive condition feature correlation dataset to each first centroid, and to determine the set to which each data point in the adhesive condition feature correlation dataset belongs;

[0035] After all the data is aggregated into the corresponding sets, the centroid of each set is recalculated and used as the second centroid.

[0036] Calculate the distance between the first centroid and the second centroid, and determine whether the algorithm should terminate based on the calculated distance. If the distance between the second centroid and the first centroid is less than a first preset threshold, the algorithm terminates; otherwise, repeat the above steps to cluster the data.

[0037] Furthermore, using the Euclidean distance calculation method, the distance from each data point in the adhesion condition feature correlation dataset to each first centroid is calculated, determining the set to which each data point in the adhesion condition feature correlation dataset belongs, including determining the pseudo-nearest neighbor data points of each first centroid, including...

[0038] Given any first centroid d in the adhesion condition feature correlation dataset m Select the first centroid d m A neighboring data point d j According to the Euclidean distance calculation method, the first centroid d m and neighboring data points d j The following conditions must be met:

[0039] In the formula, v m and v j The first centroid d of the system output m and neighboring data points d j The position, o(·) 2 N represents higher-order terms of the system. c Let m represent the number of data points in the data set related to the characteristics of the adhesive working condition, m represent the m-th first centroid in the data set related to the characteristics of the adhesive working condition, and j represent the j-th data point in the data set related to the characteristics of the adhesive working condition.

[0040] If data point d j The first mass d m The pseudo-nearest neighbor data points then satisfy:

[0041] In the formula, R m For the model function at data point d m The 2-norm of the Jacobian matrix at the given location;

[0042] If data point d j The first mass d m If the pseudo-nearest neighbor data points are d, then the data point d will be... j Aggregate to the first mass center d m The set it belongs to.

[0043] Furthermore, determining the set to which each data point in the adhesion condition feature-related dataset belongs also includes determining the number R of first centroids. m ,include,

[0044] Determine the clustering optimization model:

[0045] In the formula, Q(R) m N is the proportion of data points in the total data points in the model input space where the first centroid has a pseudo-nearest neighbor. rN is the number of pseudo-nearest neighbors of the first centroid. c The number of data points in the dataset related to adhesion condition characteristics;

[0046] Each cluster with its first centroid can be considered a local linearization within the model input space, satisfying:

[0047] In the formula, and b i For the i-th cluster d m The linearization parameters;

[0048] Calculate the cluster membership of the dataset related to adhesion condition features, satisfying:

[0049] In the formula, h i，m The first mass d m The membership degree D in the i-th cluster i,m and D q，m The first mass center d m The distances to the i-th and q-th cluster centers, where w∈[1,∞) is the weighting exponent;

[0050] Let data point d m The center of the i-th cluster is c i Determine the fuzzy partitioning matrix of the i-th cluster. and the fuzzy covariance matrix Ψ i The smallest eigenvector in,

[0051] In the formula, p and v represent the fuzzy covariance matrix Ψ i Different minimum eigenvalue vectors

[0052] From formula (1.9), we can derive:

[0053] in,

[0054] Based on formulas (1.9), (1.10), and (1.11), we can obtain:

[0055] Based on formulas (1.5), (1.9), (1.10), and (1.12), we can obtain:

[0056] Furthermore, R m The first centroid represents R. m There are N sets. z={1, 2, ..., R} m}

[0057] Furthermore, based on the number N of train adhesion condition levels z Based on the current operating status data, identify the current adhesion condition level N of the train. d include,

[0058] Based on the current operating status data, determine the current traction condition, the current adhesion condition characteristic dataset, and the current speed measurement deviation dataset E′. r,j ;

[0059] Based on the determined current traction condition, current adhesion condition characteristic related dataset, and current speed measurement deviation dataset E′ r,j And the number N of train adhesion condition levels z Principal component analysis of data features is used to determine the current adhesion condition level N. d , where N d ∈N z .

[0060] Furthermore, principal component analysis of data features is used to determine the current adhesion condition level N. d include,

[0061] Step S101: Standardize the current adhesion condition feature-related dataset D(t) to obtain the standardization matrix D. s (t);

[0062] Step S102: Solve for the normalized matrix D s The covariance matrix XX of (t) T ;

[0063] Step S103: For the covariance matrix XX T Eigenvalue decomposition is performed to obtain eigenvalues ​​and corresponding eigenvectors, where the eigenvalues ​​represent the characteristics of the current adhesion condition in N. z The variance of each adhesion condition level in the model is represented by the eigenvectors, which represent the weights of the corresponding feature directions.

[0064] Step S104: Sort the feature values ​​from largest to smallest, and select the sticky feature corresponding to the largest feature value k as the principal component;

[0065] Step S105: Perform a comprehensive evaluation of the principal components determined in step S104, and compare the evaluation results with N. z The centroids in each set of the various levels are compared. If the evaluation result is closest to the evaluation index of one of the centroids, then the train adhesion condition level corresponding to that centroid is the current train adhesion condition level N. d .

[0066] Furthermore, input the current adhesion condition level N of the train. d Establishing the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level includes,

[0067] Determine the current train adhesion condition level N d The following adhesion coefficient:

[0068] μ s，1 =Φ(N d (2-1)

[0069] In the formula, μ s，1 N represents the adhesion coefficient under the current adhesion condition level. d N represents the level of the current adhesion condition. d ∈N z ;

[0070] The adhesion condition levels are sorted in order of adhesion status from poorest to best. The mapping function Φ between the current adhesion condition level and the adhesion coefficient satisfies:

[0071] In the formula, C represents the proportional reduction factor for calculating the adhesion grade, and α represents the safety reduction factor;

[0072] Calculate the adhesion coefficient under the current train operating conditions:

[0073] In the formula, μ s，2 Here, F is the calculated value of the adhesion coefficient, M is the traction or braking force of the train, and g is the acceleration due to gravity.

[0074] The adhesion coefficient under the current train operating conditions determines the current train adhesion condition level N. d The adhesion coefficient is corrected, and the current train adhesion condition level N is determined. d The following adhesion coefficients satisfy:

[0075] μ s =min(μ s，1 μ s，2 (2-4)

[0076] Current train adhesion condition level N d The mapping relationship between the adhesion coefficient and the maximum deceleration of the train under braking satisfies:

[0077] a μ =μ s g (2-5)

[0078] In the formula, a μLet g be the maximum deceleration of the train during braking, and g be the acceleration due to gravity.

[0079] Furthermore, based on the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level, the train protection curve is generated, including:

[0080] Based on the mobile authorization information, the train stopping point information is determined, including the target location s. end and target speed v end And v end =0;

[0081] The train's operating speed is divided into multiple speed segments;

[0082] Get real-time train information;

[0083] Starting from the target location, calculate the corresponding protection curve for each speed segment in the opposite direction of the train's running direction;

[0084] Based on the protection curve for each speed range, the train protection curve is output.

[0085] Furthermore, the real-time train information includes the position s of the train's front end. start Train speed v0, train mass M, train emergency braking characteristic curve Q E .

[0086] Furthermore, starting from the target location, the corresponding protection curves are calculated sequentially in each speed segment in the opposite direction of train travel, including...

[0087] Step S201: Initialize the velocity segment of the protection curve to be calculated, denoted as the i-th velocity segment. Let the final velocity and end position of the i-th velocity segment be v, respectively. end，i =v end s end，i =s end ;

[0088] Step S202: Read the initial velocity v of the i-th velocity segment. start，i ;

[0089] Step S203: Obtain the train deceleration in the i-th speed segment, and based on the initial velocity v start，i v end,i And the train's deceleration, to generate the running length of the i-th speed segment;

[0090] Step S204: Under the most unfavorable slope conditions, obtain the comprehensive deceleration of the i-th velocity segment;

[0091] Step S205: Based on the comprehensive deceleration, recalculate the running length and entry position of the i-th speed segment, and generate the protection curve in the i-th speed segment;

[0092] Step S206: Determine whether the entrance position of the i-th speed segment exceeds the train's starting position s. start If the speed limit is exceeded, then the train speed v0 is used as the upper limit of the i-th speed segment, and steps S203-S205 are executed again; if the speed limit is not exceeded, then it is determined whether the i-th speed segment is the last speed segment. If so, the algorithm ends, and the current speed limit is used as the initial position s of the train. start to the entry position s of the i-th velocity segment start，i The protection curve; if not, then start calculating the next speed segment, let i = i + 1, v end，i =v start，i-1 s end，i =s start，i-1 Repeat steps S202-S205.

[0093] Furthermore, the train's deceleration during the speed range is obtained, and based on the initial velocity v start，i v end，i In addition to the train's deceleration, the length of the i-th speed segment is generated, including:

[0094] According to the train emergency braking characteristic curve Q E Read the train braking deceleration a corresponding to the i-th speed segment 1，i ;

[0095] Based on the current adhesion condition level N of the train d The following vehicles have a maximum braking deceleration a μ The train braking deceleration a 1，i Make corrections to generate the train deceleration 'a' for the i-th speed segment. 2，i :

[0096] a 2，i =min(a 1，i a μ (2-6)

[0097] Based on the initial velocity v start，i v end，i And the train's deceleration, generating the running length of the i-th speed segment:

[0098] Where A1 is the first conversion unit coefficient.

[0099] Furthermore, under the most unfavorable slope conditions, the comprehensive deceleration for the i-th velocity segment is obtained, including:

[0100] Selected by s end，i Subtract s i，1 The position after that to position s end，i The maximum downhill slope between them is denoted by the slope value w. i When there is no downhill slope, the slope value w is recorded. i =0;

[0101] train deceleration a 2，i Make corrections to obtain the combined deceleration for the i-th velocity segment:

[0102] A2 is the second conversion unit coefficient.

[0103] Furthermore, based on the comprehensive deceleration, the running length and entry position of the i-th speed segment are recalculated, and the protection curve in the i-th speed segment is generated, including...

[0104] The length of the i-th velocity segment satisfies:

[0105] The entry position of the i-th velocity segment satisfies:

[0106] s start，i =s end，i -s i (2-10) The protection curve in the i-th speed segment is:

[0107] In the formula, x is the distance s start，i Length, s start，i ≤x≤s i v start，i Let a be the initial velocity of the i-th velocity segment. i Let be the combined deceleration of the i-th velocity segment.

[0108] Furthermore, it also includes calculating the train braking distance L. b L b It is the sum of the running lengths corresponding to multiple speed segments.

[0109] Another objective of this invention is to provide an adaptive control system based on train operating condition identification, comprising:

[0110] A multi-sensor fusion adhesion feature state self-calibration observer is used to acquire train operating status data under different traction conditions;

[0111] The data calculation module is used to calculate traction and braking performance data based on the acquired train operating status data under different traction conditions.

[0112] The data integration module is used to combine traction and braking performance data, track information data, and operation control information to form a multi-sensor fusion adhesive condition feature-related dataset.

[0113] The clustering module is used to divide the train adhesion conditions into N categories based on the dataset related to adhesion condition features. z There are several levels, of which N z It is a natural number;

[0114] The identification module is used to identify the number N of train adhesion condition levels. z Based on the current operating status data, identify the current adhesion condition level N of the train. d N d N is a natural number, and d ≤N z ;

[0115] The mapping relationship establishment module is used to input the current adhesion condition level N of the train. d Establish the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level;

[0116] The curve generation module is used to generate train protection curves based on the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level.

[0117] Furthermore, the calculated traction and braking performance data includes inputting the train's operating state data under different traction conditions into the corresponding train dynamics model. These different traction conditions include starting, acceleration, cruising, coasting, and braking phases. The train dynamics model under these different traction conditions satisfies the following:

[0118]

[0119] Where Y(t), X(t-1), and Z(t) are the model's output at time t, its internal state at time t-1, and its input at time t, respectively, and f i (X(t-1), Z(t)) represents the dynamic characteristics of the i-th traction condition at time t fitted using a deep learning algorithm, w i (t) represents the fitting error of the i-th traction condition at time t, where i≤5 and is an integer.

[0120] Furthermore, the multi-sensor fusion adhesive feature state self-calibration observer includes,

[0121] Wheel axle speed sensors and Doppler radar speed sensors are used to measure train speed;

[0122] Accelerometer sensor, used to obtain train acceleration;

[0123] Humidity sensors and temperature sensors are used to acquire the temperature and humidity in the environment, respectively.

[0124] Furthermore, the acquisition of train operating status data under different traction conditions using a multi-sensor fusion-based adhesion feature state self-calibration observer also includes:

[0125] Based on the measured train operating speed, speed test dataset V was obtained. r,j and V d,j , where V r,j V is the dataset corresponding to the train running speed measured by the wheel axle speed sensor. d,j This is a dataset corresponding to train speeds measured by Doppler radar speed sensors.

[0126] Based on dataset V r,j and V d，j The speed measurement deviation dataset E was calculated. r，j and train speed measurement dataset V c，j It can be described as follows:

[0127] V c，j =·r1*g(V r，j )+·r2*g(V d，j (1.2)

[0128] E r，j =·h(V r，j -V d，j (1.3)

[0129] In the formula, r1 and r2 are the self-adjusting weights for velocity fusion, and g(·) and h(·) represent the fusion strategy functions;

[0130] The route information data includes the route longitudinal profile data;

[0131] The operational control information includes train control commands and temporary speed limits.

[0132] Furthermore, based on the relevant dataset of adhesion condition characteristics, the train adhesion condition is divided into N z The levels include using a pseudo-nearest neighbor algorithm based on fuzzy clustering to divide the train adhesion condition into N levels. z The levels specifically include:

[0133] Based on the traction and braking performance data output from the train dynamics model, an R is estimated. m value;

[0134] R was randomly selected from the dataset related to adhesion condition features. m One data point is used as the first centroid;

[0135] The Euclidean distance method is used to calculate the distance from each data point in the adhesion condition feature correlation dataset to each first centroid, thereby determining the set to which each data point in the adhesion condition feature correlation dataset belongs; wherein,

[0136] Given any first centroid d in the adhesion condition feature correlation dataset m Select the first centroid d m A neighboring data point d j According to the Euclidean distance calculation method, the first centroid d m and neighboring data points d j The following conditions must be met:

[0137] In the formula, v m and v j The first centroid d of the system output m and neighboring data points d j The position, o(·) 2 N represents higher-order terms of the system. c Let m represent the number of data points in the data set related to the characteristics of the adhesive working condition, m represent the m-th first centroid in the data set related to the characteristics of the adhesive working condition, and j represent the j-th data point in the data set related to the characteristics of the adhesive working condition.

[0138] If data point d j For the first mass d m The pseudo-nearest neighbor data points then satisfy:

[0139] In the formula, R m For the model function at data point d m The 2-norm of the Jacobian matrix at the given location;

[0140] If data point d j For the first mass d m If the pseudo-nearest neighbor data points are d, then the data point d will be... j Aggregate to the first mass center d m The set it belongs to;

[0141] After all the data is aggregated into the corresponding sets, the centroid of each set is recalculated and used as the second centroid.

[0142] Calculate the distance between the first centroid and the second centroid, and determine whether the algorithm should terminate based on the calculated distance. If the distance between the second centroid and the first centroid is less than a first preset threshold, the algorithm terminates; otherwise, repeat the above steps to cluster the data.

[0143] Furthermore, determining the set to which each data point in the adhesion condition feature-related dataset belongs also includes determining the number R of first centroids. m,include,

[0144] Determine the clustering optimization model:

[0145] In the formula, Q(R) m N is the proportion of data points in the total data points in the model input space where the first centroid has a pseudo-nearest neighbor. r N is the number of pseudo-nearest neighbors of the first centroid. c The number of data points in the dataset related to adhesion condition characteristics;

[0146] Each cluster with its first centroid can be considered a local linearization within the model input space, satisfying:

[0147] In the formula, and b i For the i-th cluster d m The linearization parameters;

[0148] Calculate the cluster membership of the dataset related to adhesion condition features, satisfying:

[0149] In the formula, h i，m The first mass d m The membership degree D in the i-th cluster i，m and D q，m The first mass center d m The distances to the i-th and q-th cluster centers, where w∈[1,∞) is the weighting exponent;

[0150] Let data point d m The center of the i-th cluster is c i Determine the fuzzy partitioning matrix of the i-th cluster. and the fuzzy covariance matrix Ψ i The smallest eigenvector in,

[0151] In the formula, p and v represent the fuzzy covariance matrix Ψ i Different minimum eigenvalue vectors

[0152] From formula (1.9), we can derive:

[0153] in,

[0154] Based on formulas (1.9), (1.10), and (1.11), we can obtain:

[0155] Based on formulas (1.5), (1.9), (1.10), and (1.12), we can obtain:

[0156] Among them, R m The first centroid represents R. m There are N sets. z ={1, 2, ..., R} m}

[0157] Furthermore, based on the number N of train adhesion condition levels z Based on the current operating status data, identify the current adhesion condition level N of the train. d include,

[0158] Based on the current operating status data, determine the current traction condition, the current adhesion condition characteristic dataset, and the current speed measurement deviation dataset E′. r，j ;

[0159] Based on the determined current traction condition, current adhesion condition characteristic related dataset, and current speed measurement deviation dataset E′ r，j And the number N of train adhesion condition levels z Principal component analysis of data features is used to determine the current adhesion condition level N. d , where N d ∈N z Specifically, including,

[0160] Step S101: Standardize the current adhesion condition feature-related dataset D(t) to obtain the standardization matrix D. s (t);

[0161] Step S102: Solve for the normalized matrix D s The covariance matrix XX of (t) T ;

[0162] Step S103: For the covariance matrix XX T Eigenvalue decomposition is performed to obtain eigenvalues ​​and corresponding eigenvectors, where the eigenvalues ​​represent the characteristics of the current adhesion condition in N. z The variance of each adhesion condition level in the model is represented by the eigenvectors, which represent the weights of the corresponding feature directions.

[0163] Step S104: Sort the feature values ​​from largest to smallest, and select the sticky feature corresponding to the largest feature value k as the principal component;

[0164] Step S105: Perform a comprehensive evaluation of the principal components determined in step S104, and compare the evaluation results with N. zThe centroids in each set of the various levels are compared. If the evaluation result is closest to the evaluation index of one of the centroids, then the train adhesion condition level corresponding to that centroid is the current train adhesion condition level N. d .

[0165] Furthermore, input the current adhesion condition level N of the train. d Establishing the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level includes,

[0166] Determine the current train adhesion condition level N d The following adhesion coefficient:

[0167] μ s，1 =Φ(N d (2-1)

[0168] In the formula, μ s，1 N represents the adhesion coefficient under the current adhesion condition level. d N represents the level of the current adhesion condition. d ∈N z ;

[0169] The adhesion condition levels are sorted in order of adhesion status from poorest to best. The mapping function Φ between the current adhesion condition level and the adhesion coefficient satisfies:

[0170] In the formula, α represents the safety reduction factor, and C represents the proportional reduction factor for calculating the adhesion level;

[0171] Calculate the adhesion coefficient under the current train operating conditions:

[0172] In the formula, μ s，2 Here, F is the calculated value of the adhesion coefficient, M is the traction or braking force of the train, and g is the acceleration due to gravity.

[0173] The adhesion coefficient under the current train operating conditions determines the current train adhesion condition level N. d The adhesion coefficient is corrected, and the current train adhesion condition level N is determined. d The following adhesion coefficients satisfy:

[0174] μ s =min(μ s，1 μ s，2 (2-4)

[0175] Current train adhesion condition level N d The mapping relationship between the adhesion coefficient and the maximum deceleration of the train under braking satisfies:

[0176] a μ =μ s g (2-5)

[0177] In the formula, a μ Let g be the maximum deceleration of the train during braking, and g be the acceleration due to gravity.

[0178] Furthermore, based on the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level, the train protection curve is generated, including:

[0179] Based on the mobile authorization information, the train stopping point information is determined, including the target location s. end and target speed v end And v end =0;

[0180] The train's operating speed is divided into multiple speed segments;

[0181] Obtain real-time train information, including the position s of the train's front end. start Train speed v0, train mass M, train emergency braking characteristic curve Q E ;

[0182] Starting from the target location, calculate the corresponding protection curve for each speed segment in the opposite direction of the train's running direction;

[0183] Based on the protection curve for each speed range, the train protection curve is output.

[0184] Furthermore, starting from the target location, the corresponding protection curves are calculated sequentially in each speed segment in the opposite direction of train travel, including...

[0185] Step S201: Initialize the velocity segment of the protection curve to be calculated, denoted as the i-th velocity segment. Let the final velocity and end position of the i-th velocity segment be v, respectively. end，i =v end s end，i =s end ;

[0186] Step S202: Read the initial velocity v of the i-th velocity segment. start，i ;

[0187] Step S203: Obtain the train deceleration in the i-th speed segment, and based on the initial velocity v start，i v end，i And the train's deceleration, generating the running length of the i-th speed segment; including,

[0188] According to the train emergency braking characteristic curve QE Read the train braking deceleration a corresponding to the i-th speed segment 1，i ;

[0189] Based on the current adhesion condition level N of the train d The following vehicles have a maximum braking deceleration a μ The train braking deceleration is corrected to generate the train running deceleration 'a' for the i-th speed segment. 2，i :

[0190] a 2，i =min(a 1，i a μ (2-6)

[0191] Based on the initial velocity v start，i v end，i And the train's deceleration, generating the running length of the i-th speed segment:

[0192] Where A1 is the first conversion unit coefficient;

[0193] Step S204: Under the most unfavorable slope conditions, obtain the comprehensive deceleration of the i-th velocity segment, including,

[0194] Selected by s end，i Subtract s i，1 The position after that to position s end，i The maximum downhill slope between them is denoted by the slope value w. i When there is no downhill slope, the slope value w is recorded. i =0;

[0195] train deceleration a 2，i Make corrections to obtain the combined deceleration for the i-th velocity segment:

[0196] Where A2 is the second conversion unit coefficient;

[0197] Step S205: Based on the comprehensive deceleration, recalculate the running length and entry position of the i-th speed segment, and generate the protection curve in the i-th speed segment, including...

[0198] The length of the i-th velocity segment satisfies:

[0199] The entry position of the i-th velocity segment satisfies:

[0200] s start，i =s end，i -s i (2-10)

[0201] The protection curve for the i-th speed segment is:

[0202] In the formula, x is the distance s start,i Length, s start,i ≤x≤s i v start,i Let a be the initial velocity of the i-th velocity segment; i Let be the combined deceleration of the i-th velocity segment;

[0203] Step S206: Determine whether the entrance position of the i-th speed segment exceeds the train's starting position s. start If the speed limit is exceeded, then the train speed v0 is used as the upper limit of the i-th speed segment, and steps S203-S205 are executed again; if the speed limit is not exceeded, then it is determined whether the i-th speed segment is the last speed segment. If so, the algorithm ends, and the current speed limit is used as the initial position s of the train. start to the entry position s of the i-th velocity segment start，i The protection curve; if not, then start calculating the next speed segment, let i = i + 1, v end，i =v start，i-1 s end,i =s start,i-1 Repeat steps S202-S205.

[0204] The adaptive control method of this invention uses a clustering algorithm to cluster and classify adhesion conditions, making the analysis of adhesion conditions during train operation more accurate. Based on the clustering analysis, the current adhesion condition level is obtained, and a train protection curve is adaptively generated to protect the train operation, making the control of the train more reliable and improving the safety of train operation.

[0205] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0206] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0207] Figure 1 shows a schematic flowchart of an adaptive control method based on train operation condition identification in an embodiment of the present invention;

[0208] Figure 2 shows a structural diagram of an adaptive control system based on train operation condition identification in an embodiment of the present invention. Detailed Implementation

[0209] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0210] As shown in Figure 1, this embodiment of the invention introduces an adaptive control method based on train operating condition identification. The control method includes: first, acquiring train operating state data under train traction conditions using a multi-sensor fusion adhesion feature state self-calibration observer, and calculating traction and braking performance data; then, combining the traction and braking performance data, track information data, and operation control information to form a multi-sensor fusion adhesion condition feature related dataset; and finally, based on the adhesion condition feature related dataset, dividing the train adhesion conditions into N... z There are several levels, of which N z It is a natural number; then, based on the number N of train adhesion condition levels. z Based on the current operating status data, identify the current adhesion condition level N of the train. d N d N is a natural number, and d ≤N z Then, enter the current adhesion level N of the train. d This invention establishes a mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level. Finally, based on this mapping relationship, a train protection curve is generated. The adaptive control method of this invention uses a clustering algorithm to classify and categorize adhesion conditions, making the analysis of adhesion conditions during train operation more accurate. Then, based on the clustering analysis, the adhesion condition level is obtained, and a train protection curve is adaptively generated, making train control more reliable and improving the safety of train operation.

[0211] Specifically, the calculated traction and braking performance data includes inputting the train's operating state data under different traction conditions into the corresponding train dynamics model to output traction and braking performance data. The different train traction conditions include five operating stages: starting, acceleration, cruising, coasting, and braking. The train dynamics model under different traction conditions satisfies the following:

[0212]

[0213] Where Y(t), X(t-1), and Z(t) are the model's output at time t, its internal state at time t-1, and its input at time t, respectively, and f i (X(t-1), Z(t)) represent the dynamic characteristics of the i-th traction condition at time t fitted using a deep learning algorithm, w i (t) represents the fitting error of the i-th traction condition at time t, where i = 1, 2, ..., 5, i.e., i ≤ 5, and is a positive integer. Further, a multi-sensor fusion adhesion characteristic state self-calibration observer is used to acquire train operating status data under different train traction conditions. This includes measuring train speed using wheel axle speed sensors and Doppler radar speed sensors; acquiring train acceleration using accelerometer sensors; and acquiring ambient temperature and humidity using humidity and temperature sensors, respectively. Based on the multi-sensor fusion data of the train under different traction conditions, real-time observation of train adhesion condition characteristic information is achieved. However, this is not limited to this; each operating status data can be observed using multiple sensors with the same measurement target but different categories, and calibration selection can be performed to obtain the most comprehensive and accurate operating data. For example, during the startup phase, if the speed data collected by the wheel axle speed sensor reflects the real-time operating status better than the speed data collected by the Doppler radar speed sensor, then the speed data collected by the wheel axle speed sensor is used. Therefore, the self-calibrating observer of adhesion characteristics using multi-sensor fusion provides a more comprehensive observation of wheel adhesion, thus improving the accuracy of adhesion condition observation.

[0214] In this embodiment of the invention, the method of acquiring train operating status data under different traction conditions using a multi-sensor fusion adhesion feature state self-calibration observer further includes:

[0215] Based on the measured train operating speed, speed test dataset V was obtained. r,j and V d,j , where V r,j V is the dataset corresponding to the train running speed measured by the wheel axle speed sensor. d,j This is a dataset corresponding to train speeds measured by Doppler radar speed sensors.

[0216] Based on dataset V r,j and V d,j The speed measurement deviation dataset E was calculated. r，j and train speed measurement dataset V c，j :

[0217] V c，j =·r1*g(V r，j )+·r2*g(V d，j(1.2)

[0218] E r，j =·h(V r，j -V d，j (1.3)

[0219] In the formula, r1 and r2 are the self-adjusting weights for velocity fusion, g(·) and h(·) represent the fusion strategy functions, and V r，j V d,j V c，j The subscripts r,j, d,j, and c,j in E are used to distinguish different datasets. Similarly, E r，j The subscripts r and j in the table are also used to distinguish datasets. j represents a data point in the dataset, that is, the above datasets are all sets of data points.

[0220] Furthermore, traction and braking performance data, track information data, and operation control information are combined to form a multi-sensor fusion dataset related to adhesion condition characteristics. The track information data includes track longitudinal profile data; the operation control information includes train control commands and temporary speed limit information. The output of the multi-sensor fusion observer, combined with the track information data and operation control information, serves as an offline dataset for classifying adhesion condition characteristic levels, making adhesion condition analysis more reliable.

[0221] Cluster analysis, or clustering, is a method of statically classifying similar objects into different groups or subsets, ensuring that members within the same subset share similar attributes. In this embodiment of the invention, based on a dataset related to adhesion characteristics, a fuzzy clustering-based False Nearest Neighbors (FC-FNN) algorithm is used to analyze and calculate the number N of train adhesion condition levels. z , where N z ={1, 2, ..., R} m}, R m This indicates the number of clusters combining train adhesion feature data, specifically including:

[0222] First, based on the traction and braking performance data output from the train dynamics model, an R is estimated. m Value; specifically, choose an R based on experience. m value.

[0223] Secondly, R is randomly selected from the dataset related to adhesion condition features. m 1. Data points are used as the first centroid, where R... m The whole represents a numerical value; the subscript 'm' is used to indicate that R... mThese are values ​​related to the center of mass;

[0224] Then, using the Euclidean distance calculation method, the distance from each data point in the adhesion condition feature correlation dataset to each first centroid is calculated, thus determining the set to which each data point in the adhesion condition feature correlation dataset belongs; specifically, the clustering analysis process includes: given any first centroid d in the adhesion condition feature correlation dataset... m Select the first centroid d m A neighboring data point d j According to the Euclidean distance calculation method, the first centroid d m and neighboring data points d j The following conditions must be met:

[0225] In the formula, v m and v j The first centroid d of the system output m and neighboring data points d j The position, o(·) 2 N represents higher-order terms of the system. c Let m be the number of data points in the adhesive condition feature correlation dataset, and m represent the m-th first centroid in the adhesive condition feature correlation dataset, where m∈R. m , j represents the j-th data point in the adhesive condition feature related dataset, and m and j are both integers greater than 0;

[0226] If data point d j The first mass d m The pseudo-nearest neighbor data points can be obtained according to Cauchy's inequality and formula (1.4):

[0227] In the formula, R m For the model function at data point d m The 2-norm of the Jacobian matrix at the given location;

[0228] If data point d j The first mass d m If the pseudo-nearest neighbor data points are d, then the data point d will be... j Aggregate to the first mass center d m The set it belongs to.

[0229] Determining the clustering optimization model, i.e., the model order estimation problem, can be transformed into the following optimization problem:

[0230] In the formula, Q(R) m N is the proportion of data points in the total data points in the model input space where the first centroid has a pseudo-nearest neighbor. r N is the number of data points with pseudo-nearest neighbors to the first centroid.c R is the number of data points in the dataset related to the adhesion condition characteristics; R in the FC-FNN algorithm shown in Equation (1.5) m Setting a threshold is one of the key steps in the problem of sticky feature type estimation.

[0231] Each cluster with its first centroid can be considered a local linearization within the model input space, satisfying:

[0232] In the formula, and b i For the i-th cluster d m The linearization parameters;

[0233] The cluster membership of the dataset related to adhesion condition features is calculated. Specifically, the Gath-Geva clustering method (a probability-based distance clustering algorithm) is used to calculate the cluster membership of the dataset related to adhesion condition features.

[0234] In the formula, h i，m The first mass d m The membership degree D in the i-th cluster i，m and D q，m The first mass center d m The distances to the i-th and q-th cluster centers, where w∈[1,∞) is the weighting exponent;

[0235] Let data point d m The center of the i-th cluster is c i Determine the fuzzy partitioning matrix of the i-th cluster. and the fuzzy covariance matrix Ψ i The smallest eigenvector in,

[0236] In the formula, p and v represent the fuzzy covariance matrix Ψ i Different minimum eigenvalue vectors

[0237] From formula (1.9), we can derive:

[0238] in,

[0239] Based on formulas (1.9), (1.10), and (1.11), we can obtain:

[0240] Based on formulas (1.5), (1.9), (1.10), and (1.12), we can obtain:

[0241] Therefore, by combining formula (1.13), the number of clusters (i.e., the number of datasets) R of train adhesion feature data can be calculated. m The FC-FNN algorithm is used to cluster each data point in the dataset related to adhesion condition features. This method is efficient, usable, and converges quickly. Furthermore, it yields the number N of adhesion feature categories (i.e., the number of train adhesion condition levels) under complex and variable environments. z This allows for the classification of adhesion conditions, improving analysis efficiency.

[0242] Then, after all the data is aggregated into the corresponding sets, the centroid of each set is recalculated and used as the second centroid.

[0243] Calculate the distance between the first and second centroids for each set, and determine whether the algorithm terminates based on the calculated distance. If the distance between the second and first centroids is less than a first preset threshold, the clustering has achieved the desired result, and the algorithm terminates. Otherwise, iterate the above steps to cluster the data. The first preset threshold can be 0.1, 0.5, etc., but is not limited to these; 0.3, etc., is also applicable to this invention. Further, R... m The sets themselves are classified based on different adhesion conditions, that is, divided into N... z The adhesion condition level of each train, namely N z It is also a set, and the set contains R. m Number, N z ={1, 2, ..., R} m}, R m This indicates the number of clusters in the train adhesion feature data. It should be noted that each data point in the adhesion condition feature-related dataset in this embodiment of the invention is different. For example, radar speed measurement is a non-contact measurement method and is not affected by adhesion conditions, while the measurement results of wheel axle speed sensors are easily affected by adhesion conditions. Under conditions of acceleration slippage and deceleration slippage, there are deviations between radar speed measurement and wheel axle speed measurement results. Furthermore, the multi-sensor fusion deviation is significantly different under train operating states such as low-speed operation, high-speed operation, and coasting. Therefore, each data point in the adhesion condition feature-related dataset reflects the adhesion state. During the clustering process, data points with better adhesion states will mainly cluster in the same set. Thus, different datasets, based on the different adhesion conditions of the data, constitute datasets with different adhesion condition levels. For example, if R... m =5, which means there are 5 clustered datasets, N z ={1, 2, 3, 4, 5}. Clustering and classifying different data points can better capture the adhesion conditions during train operation and improve the reliability of train operation.

[0244] In this embodiment of the invention, based on the number N of train adhesion condition levels z Based on the current operating status data, identify the current adhesion condition level N of the train. d include,

[0245] First, based on the current operating status data, determine the current traction condition, the current adhesion condition characteristic dataset, and the current speed measurement deviation dataset E′. r，j Among them, the current adhesion condition characteristic related dataset and the current velocity measurement deviation dataset E′ r，j The acquisition method is the same as the dataset related to the adhesion condition characteristics and the velocity measurement deviation dataset E under different traction conditions mentioned above. r，j Similarly, this only solves for the relevant datasets under the current running state, and the specifics will not be elaborated further. In addition, the current time range can be a cycle of 10 seconds, that is, the current running data is acquired and the corresponding dataset is calculated once every 10 seconds, but it is not limited to this. It can also be 15 seconds, 1 minute, etc., which are also applicable to this invention.

[0246] Based on the determined current traction condition, current adhesion condition characteristic related dataset, and current speed measurement deviation dataset E′ r，j And the number N of train adhesion condition levels z Principal Component Analysis (PCA) is used to determine the current adhesion condition level N of the train. d N d N is a natural number, and d ≤N z , where N d ∈N z The subscripts d and z are only used to distinguish N. d N z This has no special meaning. The specific algorithm steps are as follows:

[0247] Step S101: Standardize the current adhesion condition feature-related dataset D(t) (i.e., the adhesion condition feature-related dataset at time t) to obtain the standardization matrix D. s (t);

[0248] Step S102: Solve for the normalized matrix D s The covariance matrix XX of (t) T ;

[0249] Step S103: For the covariance matrix XX T Eigenvalue decomposition is performed to obtain eigenvalues ​​and corresponding eigenvectors, where the eigenvalues ​​represent the characteristics of the current adhesion condition in N. zThe variance of each adhesion condition level (characteristic direction) in the model is represented by the eigenvector, which represents the weight of the corresponding characteristic direction.

[0250] Step S104: Sort the feature values ​​from largest to smallest, and select the sticky feature corresponding to the largest feature value k as the principal component;

[0251] Step S105: Perform a comprehensive evaluation of the principal components determined in step S104, and compare the evaluation results with N. z The centroids in each set of the various levels are compared. If the evaluation result is closest to the evaluation index of one of the centroids, then the train adhesion condition level corresponding to that centroid is the current train adhesion condition level N. d For example, if the principal component is the current velocity measurement deviation dataset E′ r,j (i.e., the measurement deviation E′ between the wheel axle speed sensor and the Doppler radar speed sensor) r,j The speed deviation in the current speed measurement deviation dataset E′ r，j A comprehensive evaluation can be performed, and the evaluation result can be the average of all speed deviations in the dataset, but it is not limited to this. Other evaluation methods are also applicable to this invention. The evaluation result is then compared with N. z The centroids in each set of the levels are compared. If the evaluation result is closest to the evaluation index of one of the centroids, then the adhesion condition level corresponding to that centroid is the current adhesion condition level N of the train. d For example, the evaluation metric that best approximates the value is: the velocity measurement deviation dataset E corresponding to one of the centroids among all centroids. r，j If the average speed deviation in the centroid has the smallest difference from the evaluation result, meaning they are closest, then the adhesion condition level corresponding to that centroid is the current adhesion condition level N of the train. d .

[0252] In this embodiment of the invention, the current adhesion condition level N of the train is input. d Establishing the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level includes:

[0253] Determine the current train adhesion condition level N d The following adhesion coefficient:

[0254] μ s,1 =Φ(N d (2-1)

[0255] In the formula, μ s,1 This represents the adhesion coefficient under the current adhesion condition level. The subscript s,1 is only used to distinguish different adhesion coefficients. The same applies to other subscripts corresponding to the adhesion coefficient below, and will not be repeated here; N d Given the current adhesion condition level, there exists N. d∈N z .

[0256] The adhesion condition levels are sorted in order of adhesion status from poorest to best. The mapping function Φ between the current adhesion condition level and the adhesion coefficient satisfies:

[0257] In the formula, α represents the safety reduction factor, and N z The number of adhesion condition levels for the train is given by C, which represents the proportional reduction factor for calculating the adhesion level. The value of C can be 0.9, but is not limited to this. Values ​​in the range of 0.5-0.9 are also applicable to this invention.

[0258] Calculate the adhesion coefficient under the current train operating conditions:

[0259] In the formula, μ s，2 Here, F is the calculated value of the adhesion coefficient, M is the traction or braking force of the train, and g is the acceleration due to gravity.

[0260] The adhesion coefficient under the current train operating conditions determines the current train adhesion condition level N. d The adhesion coefficient is corrected, and the current train adhesion condition level N is determined. d The following adhesion coefficients satisfy:

[0261] μ s =min(μ s，1 μ s，2 (2-4)

[0262] Current adhesion condition level N d The mapping relationship between the adhesion coefficient and the maximum deceleration of the train under braking satisfies:

[0263] a μ =μ s g (2-5)

[0264] In the formula, a μ The maximum deceleration of the train is given by μ, which is used to distinguish different decelerations. The same applies to other subscripts corresponding to decelerations, which will not be repeated here. g is the acceleration due to gravity.

[0265] In this embodiment of the invention, the train protection curve is generated based on the mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level.

[0266] Based on the mobile authorization information, the train stopping point information is determined, including the target location s. end and target speed v end And v end=0;

[0267] The train's operating speed is divided into multiple speed segments, and a deceleration is generated for each speed segment, i.e., the comprehensive deceleration described below. This means that the speed-limiting protection process from the train's current speed (speed limit) to the target speed of 0 is divided into multiple speed segments. The default segmentation method is a six-segment method (0km / h-60km / h, 60km / h-105km / h, 105km / h-160km / h, 160km / h-220km / h, 220km / h-265km / h, 265km / h-325km / h), but it is not limited to this and can be adjusted according to actual needs. A train deceleration is generated for the protection process in each speed segment, and this deceleration cannot exceed the minimum value of the deceleration curve corresponding to the speed segment.

[0268] Obtain real-time train information; wherein, the real-time train information includes the position s of the train's front end. start Train speed v0, train mass M, train emergency braking characteristic curve Q E ;

[0269] Starting from the target location, the corresponding protection curve is calculated sequentially for each speed segment in the opposite direction of train travel; specifically including,

[0270] Step S201: Initialize the velocity segment of the protection curve to be calculated, denoted as the i-th velocity segment. Let the final velocity and end position of the i-th velocity segment be v, respectively. end，i =v end s end，i =s end ;

[0271] Step S202: Read the initial velocity v of the i-th velocity segment. start，i ;

[0272] Step S203: Obtain the train deceleration in the i-th speed segment, and based on the initial velocity v start，i v end，i And the train's deceleration, generating the length of the i-th speed segment; specifically, based on the train's emergency braking characteristic curve Q... E Read the train braking deceleration 'a' corresponding to the speed segment. 1，i The initial and final velocities of any speed segment are the initial and final velocities known when the speed segment is divided. For example, in the speed segment of 0 km / h-60 km / h, the initial velocity is 60 km / h and the final velocity is 0 km / h.

[0273] Based on the current adhesion condition level N d The following vehicles have a maximum braking deceleration a μ Correct the train braking deceleration to generate the train running deceleration a. 2，i:

[0274] a 2，i =min(a 1，i a μ (2-6)

[0275] Based on the initial velocity v start,i v end,i and train deceleration a 2,i Generate the running length of the i-th speed segment:

[0276] Wherein, A1 is the first conversion unit coefficient, and the value of A1 can be 25.92, where 25.92 = 2 * 3.6 * 3.6, 3.6 is the conversion coefficient between km / h and m / s, and 2 is the coefficient of the kinematic equation, but it is not limited to this, and other conversion values ​​are also applicable to this invention.

[0277] Step S204: Under the most unfavorable slope conditions, obtain the comprehensive deceleration of the i-th velocity segment; specifically, select the deceleration from s end，i Subtract s i，1 The position obtained after the change is position s end，i The maximum downhill slope between them is denoted by the slope value w. i When there is no downhill slope, the slope value w is recorded. i =0;

[0278] Train deceleration a 2，i Make corrections to obtain the combined deceleration for the i-th velocity segment:

[0279] Wherein, A2 is the second conversion unit coefficient, and the value of A2 can be 102, that is, 102 = 1000 / 9.8, where 1000 is the conversion coefficient between t (ton) and kg (kilogram), and 9.8 is the gravitational acceleration, but it is not limited to this. Conversion values ​​in other cases are also applicable to this invention.

[0280] Step S205: Based on the comprehensive deceleration, recalculate the running length and entry position of the i-th speed segment, and generate the protection curve in the i-th speed segment; specifically,

[0281] The running length of the i-th speed segment satisfies:

[0282] The entry position of the i-th velocity segment satisfies:

[0283] s start，i =s end，i -s i (2-10) The protection curve in the i-th speed segment is:

[0284] In the formula, x is the distance s start，i Length, s start，i ≤x≤s i v start,i Let a be the initial velocity of the i-th velocity segment; i Let be the combined deceleration of the i-th velocity segment.

[0285] Step S206: Determine the entry position s of the i-th velocity segment. start，i Has the train exceeded its starting position s? start If the speed limit is exceeded, then the train speed v0 is used as the upper limit of the i-th speed segment, and steps S203-S205 are executed again; if the speed limit is not exceeded, then it is determined whether the i-th speed segment is the last speed segment. If so, the algorithm ends, and the current speed limit is used as the initial position s of the train. start to the entry position s of the i-th velocity segment start，i If the protection curve is not found, then start calculating the next speed segment, let i = i + 1, v end，i =v start，i-1 s end，i =s start，i-1 Then, repeat steps S202-S205. Here, the aforementioned current speed limit is the train's current speed limit value. For example, if the train is currently running at the speed limit of 350 km / h, then it will continue to run at 350 km / h until the i-th speed segment.

[0286] Then, based on the protection curve for each speed segment, the train protection curve is output, which is achieved by connecting the curves for each speed segment. Furthermore, the method also calculates the running time for each speed segment and the total braking distance L of the train. b ,in,

[0287] The running time of the i-th speed segment satisfies:

[0288] Among them, A3 is the third conversion unit coefficient. The value of A3 can be 3.6, which is the conversion coefficient between km / h and m / s, but it is not limited to this. Other conversion values ​​are also applicable to this invention.

[0289] L b It is the sum of the running lengths corresponding to multiple speed segments.

[0290] This protection curve defines the speed limit for the train at each position between its current position and the authorized movement position (target position), and can directly generate the travel time for each speed segment and the total braking distance of the train. Adaptive control of the train based on the adhesion condition identification-generated protection curve effectively solves the problems of insufficient or excessive protection in existing technologies, improving train operation safety. Furthermore, in this embodiment of the invention, the impact of real-time changes in operating conditions and train operating status can be considered, improving the accuracy and reliability of the protection curve.

[0291] As shown in Figure 2, this embodiment of the invention also introduces an adaptive control system based on train operating condition identification capable of performing the above method. This system includes a multi-sensor fusion adhesion feature state self-calibration observer, a data calculation module, a data integration module, a clustering module, an identification module, a mapping relationship establishment module, and a curve generation module. The multi-sensor fusion adhesion feature state self-calibration observer is used to acquire train operating state data under different traction conditions. The data calculation module is used to calculate traction and braking performance data based on the acquired train operating state data under different traction conditions. The data integration module is used to combine traction and braking performance data, track information data, and operation control information to form a multi-sensor fusion adhesion condition feature related dataset. The clustering module is used to divide the train adhesion conditions into N categories based on the adhesion condition feature related dataset. z There are several levels, of which N z It is a natural number; the recognition module is used to identify the number N of train adhesion condition levels. z Based on the current operating status data, identify the current adhesion condition level N of the train. d N d N is a natural number, and d ≤N z The mapping relationship establishment module is used to input the current adhesion condition level N of the train. d The system establishes a mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion condition level. The curve generation module generates a train protection curve based on this mapping relationship. Each module also executes corresponding methods consistent with those described above, which will not be repeated here. Clustering algorithms are used to classify and categorize adhesion conditions, making the analysis of adhesion conditions during train operation more accurate. Based on the adhesion condition level obtained from the cluster analysis, the system adaptively generates train protection curves, making train control more reliable and improving train operation safety.

[0292] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A self-adaptive control method based on train operation condition recognition, characterized in that, Comprising, Adopting the multi-sensor fusion adhesion characteristic state self-calibration observer to obtain the running state data of the train under different traction working conditions, and calculating the traction braking performance data; Combining the traction braking performance data, the line information data and the running control information to form a multi-sensor fusion adhesion working condition characteristic related data set; Based on the adhesion working condition characteristic related data set, the train adhesion working condition is divided into N z grades, wherein N z is a natural number; Based on the number N of train adhesion working condition levels z And the current running state data, identify the current adhesion working condition level N of the train d , N d is a natural number, and N d ≤N z ; Input the current adhesion working condition level N of the train d , and establish a mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion working condition level. Based on the mapping relationship between the adhesion coefficient under the current adhesion working condition level and the maximum deceleration of the train braking, a train protection curve is generated.

2. The self-adaptive control method based on train operation condition recognition according to claim 1, characterized in that, The calculated traction braking performance data comprises inputting the running state data of the train in different traction working conditions into corresponding train dynamics models, wherein the different traction working conditions of the train include starting, accelerating, cruising, coasting and braking stages, and the train dynamics models of the train in different traction working conditions satisfy: where Y(t), X(t-1) and Z(t) are the output of the model at time t, the internal state at time t-1 and the input at time t, respectively, f i (X(t-1), Z(t)) represents the dynamics characteristics of the i-th traction working condition at time t fitted by the deep learning algorithm, w i (t) represents the fitting error of the i-th traction working condition at time t, i≤5, And is an integer.

3. The self-adaptive control method based on train operation condition recognition according to claim 2, characterized in that, Adopting the multi-sensor fusion adhesion characteristic state self-calibration observer to obtain the running state data of the train under different traction working conditions includes, Respectively adopting the wheel shaft speed sensor and the Doppler radar speed sensor to measure the train running speed; Adopting the accelerometer sensor to obtain the train acceleration; Adopting the humidity sensor and the temperature sensor to respectively obtain the temperature and humidity in the environment.

4. The self-adaptive control method based on train operation condition recognition according to claim 3, characterized in that, Adopting the multi-sensor fusion adhesion characteristic state self-calibration observer to obtain the running state data of the train under different traction working conditions further includes, Based on the measured train running speed, the speed test data sets V r,j and V d,j are obtained respectively, wherein V r，j is a data set corresponding to the train running speed measured by the wheel axle speed sensor, and V d，j is a data set corresponding to the train running speed measured by the Doppler radar speed sensor. Based on the dataset V r，j and V d，j , the velocity measurement bias dataset E r，j and the train measurement velocity dataset V c,j can be calculated as follows: V c，j = r1*g(V r，j ) + r2*g(V d，j ) (1.2) E r，j = h(V r，j - V d，j ) (1.3) In the formula, r1 and r2 are self-adjusting weights of speed fusion, and g(·) and h(·) represent fusion strategy functions.

5. The self-adaptive control method based on train operation condition recognition according to claim 1, characterized in that, The line information data includes line longitudinal section data; The running control information includes train control instructions and temporary speed limits.

6. The self-adaptive control method based on train operation condition recognition according to claim 4, characterized in that, Based on the adhesion working condition characteristic related data set, the train adhesion working condition is divided into N z grades, including adopting a pseudo-nearest neighbor algorithm based on fuzzy clustering to divide the train adhesion working condition into N z grades, specifically including, Based on the output of the train dynamics model, the traction and braking performance data are estimated and a R m value is calculated. R data points are randomly selected from the adhesion operating condition characteristic related data set as the first centroid; m a first centroid; Adopting the Euclidean distance calculation method, the distance of each data point in the adhesion working condition characteristic related data set to each first centroid is calculated to determine the set to which each data point in the adhesion working condition characteristic related data set belongs; After all the data are collected into the corresponding set, the centroid of each set is recalculated as the second centroid; The distance between the first centroid and the second centroid is calculated, and based on the calculated distance, it is judged whether the algorithm terminates, wherein if the distance between the second centroid and the first centroid is less than a first preset threshold, the algorithm terminates, otherwise, the above steps are repeated to cluster the data. Adopting the Euclidean distance calculation method, the distance of each data point in the adhesion working condition characteristic related data set to each first centroid is calculated to determine the set to which each data point in the adhesion working condition characteristic related data set belongs includes determining the pseudo-nearest neighbor data points of each first centroid, including, 7. The adaptive control method based on train operation condition recognition according to claim 6, characterized in that, Step S104, sorting the characteristic values from large to small, and selecting the adhesion characteristic corresponding to the maximum characteristic value k as the principal component; Given any first centroid d in the adhesion condition feature correlation dataset m Select the first centroid d m A neighboring data point d j According to the Euclidean distance calculation method, the first centroid d m and neighboring data points d j The following conditions must be met: where v m and v j are the positions of the first centroid d m and the neighboring data points d j of the system, o(·) 2 denotes the higher order terms of the system, N c is the number of data points in the stick regime feature-related dataset, m denotes the mth first centroid in the stick regime feature-related dataset, and j denotes the jth data point in the stick regime feature-related dataset. If a data point d j is a pseudo-nearest neighbor data point of the first centroid d m , then the following is satisfied: where R m is the 2-norm of the Jacobian matrix of the model function at the data point d m ; If the data point d j is a pseudo-nearest neighbor data point of the first centroid d m , then the data point d j is assigned to the cluster in which the first centroid d m is located.

8. The self-adaptive control method based on train operation condition recognition according to claim 7, characterized in that, Determining the set to which each data point in the set of adhesion operating condition characteristic related data belongs further includes determining a number R of first centroids m , including, determining a clustering optimization model: wherein Q(R m ) is the proportion of the total data points in the model input space in which the first centroid exists as a pseudo-nearest neighbor data point, N r is the number of data points in which the first centroid exists as a pseudo-nearest neighbor data point, and N c is the number of data points in the data set associated with the adhesion operating condition characteristic. Each cluster of first centroids is treated as a local linearization within the model input space, satisfying: In the formulae, and b i linearization parameters for the i-th cluster d m linearization parameters for the i-th cluster d The clustering membership of the data set related to the adhesion working condition characteristics is calculated, satisfying: In the formula, h i,m The first mass d m The membership degree D in the i-th cluster i，m and D q，m The first mass center d m The distances to the i-th and q-th cluster centers, where w∈[1,∞) is the weighting exponent; Let the data points d m be located in the i-th cluster with center c i . Determine the fuzzy partition matrix U and the fuzzy covariance matrix Ψ i the minimum eigenvalue vector of wherein In the formulae, p, v represent the fuzzy covariance matrix Ψ i different minimum eigenvalue vectors From equation (1.9) it follows that: wherein Based on equation (1.9), equation (1.10) and equation (1.11), we have: Based on equation (1.5), equation (1.9), equation (1.10) and equation (1.12), we have:

9. The self-adaptive control method based on train operation condition recognition according to any one of claims 6-8, characterized in that, R m The first centroid represents a set of R m The first centroid represents a set of R z N = {1, 2, …, R m N = {1, 2, …, R 10. The adaptive control method based on train operation condition recognition according to claim 9, characterized in that, Based on the number of train adhesion working condition levels N z and current operating state data, identify the current adhesion working condition level N of the train d including, Based on the current operating state data, a current traction operating condition, a current adhesion operating condition characteristic related data set, and a current speed measurement bias data set E' are determined r,j ; based on the determined current traction operating condition, the current adhesion operating condition characteristic data set, the current speed measurement bias data set E' r,j and the number of train adhesion operating condition classes N z , using a data feature principal component analysis method to determine the current adhesion operating condition class N d , wherein N d ∈ N z .

11. The adaptive control method based on train operation condition recognition according to claim 10, characterized in that, Adopting data characteristic principal component analysis method to judge current sticking working condition grade N d including, Step S101, standardize the current sticking working condition characteristic related data set D(t) to obtain a standardized matrix D s (t); Step S102, solving the standardization matrix D s the covariance matrix XX of (t) T ; Step S103, performing eigenvalue decomposition on the covariance matrix XX T to obtain eigenvalues and corresponding eigenvectors, wherein the eigenvalues represent variances of the current adhesion working condition characteristics in N z different adhesion working condition levels, and the eigenvectors represent weights of corresponding characteristic directions. In the formula, C represents the equal reduction coefficient of adhesion level calculation, and a represents the safety reduction coefficient; Step S105, comprehensive evaluation is made on the principal components determined in step S104, and the evaluation result is compared with the centroid in each set of N z grade, if the evaluation result is closest to the evaluation index of one of the centroids, the train adhesion working condition grade corresponding to the centroid is the current train adhesion working condition grade N d .

12. The adaptive control method based on train operation condition recognition according to claim 11, characterized in that, Input the current adhesion working condition level N of the train d The mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion working condition level is established, comprising, determining the current train adhesion working condition level N d adhesion coefficient under the current train adhesion working condition level N μ s，1 = Φ(N d ) (2-1) wherein μ s，1 is the coefficient of adhesion at the current adhesion condition level; N d is the level of the current adhesion condition, N d ∈ N z ; The adhesion working condition grade categories are sequentially ordered according to adhesion states from poor to good, and a mapping relationship function Φ of the current adhesion working condition grade and the adhesion coefficient satisfies: Based on the mapping relationship between the adhesion coefficient under the current adhesion working condition level and the maximum deceleration of the train braking, a train protection curve is generated, including, calculating the adhesion coefficient in the current train operation state: In the formula, μ s，2 Here, F is the calculated value of the adhesion coefficient, M is the traction or braking force of the train, and g is the acceleration due to gravity. Based on the adhesion coefficient under the current train operation state, the adhesion working condition level N of the current train is determined d hension working condition level N of the current train is determined The adhesion coefficient is corrected, and the adhesion coefficient under the current train adhesion working condition level N d satisfies: μ s = min(μ s，1 , μ s，2 ) (2-4) The current train adhesion working condition level N d The mapping relationship of the adhesion coefficient and the maximum deceleration of train braking under the current train adhesion working condition level N satisfies: a μ = μ s g (2-5) In the formula, a μ is the maximum deceleration of the train, and g is the acceleration of gravity.

13. The adaptive control method based on train operation condition recognition according to claim 12, characterized in that, Dividing the train running speed into multiple speed segments; Based on the mobile authorization information, determine train stop point information, the train stop point information including a target position s end and a target speed v end , and v end =0. Obtaining real-time information of the train; Taking the target position as the starting point, calculating the corresponding protection curve in each speed segment in the direction opposite to the train running direction in turn; Based on the protection curve of each speed segment, the train protection curve is output. Taking the target position as the starting point, calculating the corresponding protection curve in each speed segment in the direction opposite to the train running direction in turn includes, 14. The adaptive control method based on train operation condition recognition according to claim 13, characterized in that, The train real-time information includes train head position s start , train speed v0, train mass M, train emergency braking characteristic curve Q E .

15. The adaptive control method based on train operation condition recognition according to claim 14, characterized in that, Step S204, under the most unfavorable condition of slope, the comprehensive deceleration of the i-th speed segment is obtained; Step S201, initialize the speed section of the protection curve to be calculated, denoted as the ith speed section, let the final speed and the end position of the ith speed section be: v end，i = v end , s end，i = s end ; Step S202, reading the initial speed v of the i-th speed section start，i ; In step S203, the train running deceleration of the train in the i-th speed section is obtained, and the running length of the i-th speed section is generated based on the initial speed v start，i , end，i and the train running deceleration. Step S204, under the most unfavorable condition of slope, the comprehensive deceleration of the i-th speed segment is obtained; Step S205, based on the comprehensive deceleration, recalculating the running length of the i-th speed section and the entry position of the i-th speed section, and generating a protection curve in the i-th speed section; Step S206, judging whether the entrance position of the i-th speed section exceeds the train starting position s start , if yes, taking the train speed v0 as the speed upper limit of the i-th speed section, re-executing steps S203-S205; if no, judging whether the i-th speed section is the last speed section, if yes, ending the algorithm and taking the current speed limit as the protection curve from the train starting position s start to the entrance position s start，i of the i-th speed section; if no, starting to calculate the next speed section, letting i=i+1, v end，i =v start，i-1 , s end，i =s start，i-1 , re-executing steps S202-S205.

16. The adaptive control method based on train operation condition recognition according to claim 15, characterized in that, obtaining train running deceleration of the train in the speed section, and generating the length of the i-th speed section based on the initial speed v start，i end，i and the train running deceleration,​ According to the train emergency braking characteristic curve Q E Read the train braking deceleration a corresponding to the i-th speed section 1，i ; based on the current adhesion condition level N of the train d the maximum deceleration a of the train μ the deceleration a of the train 1，i is corrected to generate the train running deceleration a of the i-th speed section 2，i : a 2，i =min(a 1，i ,a μ ) (2-6) Based on the initial speed v start，i , v end，i and the train running deceleration, the running length of the i-th speed section is generated: Wherein, A1 is the first conversion unit coefficient.

17. The adaptive control method based on train operation condition recognition according to claim 16, characterized in that, Under the most adverse conditions of slope, the comprehensive deceleration of the i-th speed section includes, Select the maximum downhill between positions s end，i Subtract s i，1 from the position to position s end，i and record the value of the slope as w i where w i = 0 when there is no downhill. The train running deceleration a 2，i The comprehensive deceleration of the i-th speed section is obtained by correction: Wherein, A2 is the second conversion unit coefficient.

18. The adaptive control method based on train operation condition recognition according to claim 17, characterized in that, Based on the comprehensive deceleration, recalculating the running length of the i-th speed section and the entry position of the i-th speed section, and generating a protection curve in the i-th speed section includes, The length of the i-th speed segment satisfies: The entry position of the i-th speed section satisfies: s start，i = s end，i -s i (2-10) The guard curve in the i-th speed segment is: In the formula, x is the distance from s start，i the length of the vehicle, s start，i ≤ x ≤ s i , v start，i is the initial speed of the i-th speed stage, a i is the comprehensive deceleration of the i-th speed stage.

19. The adaptive control method based on train operation condition recognition according to claim 18, characterized in that, Also included is calculating the train braking distance L b , L b is the sum of the operating lengths corresponding to the plurality of speed segments.

20. An adaptive control system based on train operation condition recognition, characterized in that, Including, The multi-sensor fusion adhesion characteristic state self-calibration observer is used to obtain the running state data of the train under different traction working conditions; The data calculation module is used to calculate the traction and braking performance data based on the obtained running state data of the train under different traction working conditions; The data integration module is used to combine the traction and braking performance data, the line information data and the operation control information to form a multi-sensor fusion adhesion working condition characteristic data set; The clustering module is configured to divide the train adhesion working condition into N z grades based on the adhesion working condition feature related data set, wherein N z is a natural number. The identification module is configured to identify, based on the number N of train adhesion operating condition levels z and the current operating state data, Train current adhesion working condition level N d , N d is a natural number, and N d ≤ N z ; The mapping relationship establishing module is configured to input a current adhesion working condition level N of a train d , and establish a mapping relationship between an adhesion coefficient and a maximum deceleration of train braking under the current adhesion working condition level. The curve generation module is used to generate a train protection curve based on the mapping relationship between the adhesion coefficient under the current adhesion working condition level and the maximum deceleration of the train braking.

21. The adaptive control system based on train operation condition recognition of claim 20, wherein, The calculated traction braking performance data comprises inputting the running state data of the train in different traction working conditions into corresponding train dynamics models, wherein the different traction working conditions of the train include starting, accelerating, cruising, coasting and braking stages, and the train dynamics models of the train in different traction working conditions satisfy: where Y(t), X(t-1) and Z(t) are the output of the model at time t, the internal state at time t-1 and the input at time t, respectively, f i (X(t-1), Z(t)) represents the dynamics characteristics of the i-th traction working condition at time t fitted by the deep learning algorithm, w i (t) represents the fitting error of the i-th traction working condition at time t, i≤5, and is an integer.

22. The adaptive control system based on train operation condition recognition of claim 21, wherein, The multi-sensor fusion adhesion characteristic state self-calibration observer includes, The wheel shaft speed sensor and the Doppler radar speed sensor are used to measure the train running speed; The accelerometer sensor is used to obtain the train acceleration; The humidity sensor and the temperature sensor are respectively used to obtain the temperature and humidity in the environment.

23. The adaptive control system based on train operation condition recognition of claim 22, wherein, Using the multi-sensor fusion adhesion characteristic state self-calibration observer to obtain the running state data of the train under different traction working conditions further includes, Based on the measured train running speed, the speed test data sets V r，j and V d，j are obtained respectively, wherein V r，j is a data set corresponding to the train running speed measured by the wheel axle speed sensor, and V d，j is a data set corresponding to the train running speed measured by the Doppler radar speed sensor. Based on the dataset V r，j and V d，j , the velocity measurement bias dataset E r，j and the train measurement velocity dataset V c，j can be calculated as follows: V c，j = r1*g(V r，j ) + r2*g(V d，j ) (1.2) E r，j = h(V r，j - V d，j ) (1.3) Wherein, r1 and r2 are self-adjusting weights of speed fusion, and g(·) and h(·) represent fusion strategy functions; The line information data includes line longitudinal section data; The operation control information includes train control instructions and temporary speed limits.

24. The adaptive control system based on train operation condition recognition of claim 23, wherein, Based on the adhesion working condition characteristic related data set, the train adhesion working condition is divided into N z levels, including adopting a pseudo-nearest neighbor algorithm based on fuzzy clustering to divide the train adhesion working condition into N z levels, specifically including, Based on the output of the train dynamics model, the traction and braking performance data are estimated and a R m value is calculated. R was randomly selected from the dataset related to adhesion condition features. m One data point is used as the first centroid; Using the Euclidean distance calculation method, the distance of each data point in the adhesion working condition characteristic related data set to each first centroid is calculated to determine the set to which each data point in the adhesion working condition characteristic related data set belongs; wherein, Given any first centroid d in the adhesion condition feature correlation dataset m Select the first centroid d m A neighboring data point d j According to the Euclidean distance calculation method, the first centroid d m and neighboring data points d j The following conditions must be met: where v m and v j are the positions of the first centroid d m and the neighboring data points d j of the system, o(·) 2 denotes the higher order terms of the system, N c is the number of data points in the stick regime feature related dataset, m denotes the mth first centroid in the stick regime feature related dataset, and j denotes the jth The j-th data point; If a data point d j is a pseudo-nearest neighbor data point of the first centroid d m , then the following is satisfied: where R m is the 2-norm of the Jacobian matrix of the model function at the data point d m ; If the data point d j is a pseudo-nearest neighbor data point of the first centroid d m , then the data point d j is assigned to the cluster in which the first centroid d m is located. After all the data are collected into the corresponding set, the centroid of each set is recalculated as a second centroid; The distance between the first centroid and the second centroid is calculated, and whether the algorithm terminates is judged based on the calculated distance, wherein if the distance between the second centroid and the first centroid is less than a first preset threshold, the algorithm terminates, otherwise, the above steps are repeated to cluster the data.

25. The adaptive control system based on train operation condition recognition of claim 24, wherein, Determining the set to which each data point in the set of adhesion operating condition characteristic related data belongs further includes determining a number R of first centroids m , including, determining a clustering optimization model: wherein Q(R m ) is the proportion of the total data points in the model input space that are false near neighbors of the first centroid, N r is the number of data points that are false near neighbors of the first centroid, and N c is the number of data points in the set of adhesion operating condition characteristic-related data. Each cluster of first centroids is considered as a local linearization within the model input space, satisfying: In the formulae, and b i linearization parameters for the i-th cluster d m linearization parameters for the i-th cluster d The clustering membership of the data set related to the adhesion working condition characteristics is calculated, satisfying: where h i，m is the first centroid d m is the membership in the ith cluster, D i，m and Q q,m are the distances from the first centroid d m to the ith and qth cluster centers, respectively, and w e [1,∞) is the weighting exponent. Let the data points d m be located in the i-th cluster with center c i and determine the fuzzy partition matrix U and the fuzzy covariance matrix Ψ i the minimum eigenvalue vector of Ψ wherein, In the formulae, p, v represent the fuzzy covariance matrix Ψ i different minimum eigenvalue vectors From equation (1.9) it follows that: wherein, Based on equation (1.9), equation (1.10) and equation (1.11), we have: Based on equation (1.5), equation (1.9), equation (1.10), and equation (1.12), we have: wherein R m The R m The set N z = {1, 2,..., R m}.

26. The adaptive control system based on train operation condition recognition of claim 25, wherein, Based on the number of train adhesion working condition levels N z and current operating state data, identify the current adhesion working condition level N of the train d including, Based on the current operating state data, a current traction operating condition, a current adhesion operating condition characteristic related data set, and a current speed measurement bias data set E' are determined r，j ; based on the determined current traction operating condition, the current adhesion operating condition characteristic related data set, the current speed measurement bias data set E' r，j and the number of train adhesion operating condition levels N z , using a data feature principal component analysis method to determine the current adhesion operating condition level N d , wherein N d ∈N z , specifically including, Step S101, standardize the current sticking working condition characteristic related data set D(t) to obtain a standardized matrix D s (t); Step S102, solving the standardization matrix D s the covariance matrix XX of (t) T ; Step S103, performing eigenvalue decomposition on the covariance matrix XX T to obtain eigenvalues and corresponding eigenvectors, wherein the eigenvalues represent variances of the current adhesion working condition characteristics in N z different adhesion working condition levels, and the eigenvectors represent weights of corresponding characteristic directions. Step S104, sorting the characteristic values from large to small, and selecting the adhesion characteristic corresponding to the maximum characteristic value k as the principal component; Step S105, comprehensive evaluation is made on the principal components determined in step S104, and the evaluation result is compared with the centroid in each set of N z grade, if the evaluation result is closest to the evaluation index of one of the centroids, the train adhesion working condition grade corresponding to the centroid is the current train adhesion working condition grade N d .

27. The adaptive control system based on train operation condition recognition of claim 26, wherein, Input the current adhesion working condition level N of the train d The mapping relationship between the adhesion coefficient and the maximum deceleration of the train under the current adhesion working condition level is established, comprising, Determining the current train adhesion working condition level N d adhesion coefficient: μ s，1 = Φ(N d ) (2-1) wherein μ s，1 is the coefficient of adhesion at the current adhesion condition level; N d is the level of the current adhesion condition, N d ∈ N z ; The adhesion working condition of the train is determined based on the adhesion characteristic corresponding to the maximum characteristic value k. The mapping function Φ of the grade and the adhesion coefficient satisfies: Wherein, α represents a safety reduction coefficient, and C represents a geometric reduction coefficient of adhesion level calculation; calculating the adhesion coefficient in the current train operation state: In the formula, μ s，2 Here, F is the calculated value of the adhesion coefficient, M is the traction or braking force of the train, and g is the acceleration due to gravity. The adhesion coefficient under the current train running state is corrected based on the adhesion coefficient under the current train adhesion working condition level N d The adhesion coefficient under the current train adhesion working condition level N d satisfies: μ s = min(μ s，1 , μ s，2 ) (2-4) The current train adhesion working condition level N d The mapping relationship of the adhesion coefficient and the maximum deceleration of train braking under the current train adhesion working condition level N satisfies: a μ = μ s g (2-5) In the formula, a μ is the maximum deceleration of the train, and g is the acceleration of gravity.

28. The adaptive control system based on train operation condition recognition of claim 27, wherein, Based on the mapping relationship between the adhesion coefficient under the current adhesion working condition level and the maximum deceleration of the train braking, the train protection curve is generated, including, Based on the mobile authorization information, determine train stop point information, the train stop point information including a target position s end and a target speed v end , and v end = 0. The train running speed is divided into multiple speed sections; acquiring train real-time information, the train real-time information including train head position s start , train speed v0, train mass M, train emergency braking characteristic curve Q E ; Taking the target position as a starting point, calculating a corresponding protection curve in each speed section in a direction opposite to a train running direction in sequence; Outputting a train protection curve based on the protection curve of each speed section.

29. The adaptive control system based on train operation condition recognition of claim 28, wherein, Taking the target position as a starting point, calculating a corresponding protection curve in each speed section in a direction opposite to a train running direction in sequence includes, Step S201, initialize the speed section of the protection curve to be calculated, denoted as the ith speed section, let the end speed and the end position of the ith speed section be: v end，i = v end , s end，i = s end ; Step S202, reading the initial speed v of the i-th speed section start，i ; Step S203, obtaining the train operation deceleration of the train in the i-th speed section, and generating the operation length of the i-th speed section based on the initial speed v start，i end，i and the train operation deceleration; comprising,​ According to the train emergency braking characteristic curve Q E Read the train braking deceleration a corresponding to the i-th speed section 1,i ; Based on the current adhesion condition level N of the train d The maximum deceleration a of the train μ The train braking deceleration is corrected to generate the train running deceleration a of the i-th speed section 2,i : a 2，i = min(a 1，i , a μ ) (2-6) Based on the initial speed v start，i , v end，i and the train running deceleration, the running length of the i-th speed section is generated: Wherein, A1 is a first conversion unit coefficient; Step S204, obtaining the comprehensive deceleration of the i-th speed section under the most unfavorable condition of the slope, including, Select the maximum downhill between position s end，i and position s i，1 and call the value of the slope w end，i , where w i = 0 when there is no downhill. i ​ a reduction in speed of the train a 2，i The correction is made to obtain the overall reduction in speed of the i-th speed section: Wherein, A2 is a second conversion unit coefficient; Step S205, recalculating the running length of the i-th speed section and the entry position of the i-th speed section based on the comprehensive deceleration, and generating the protection curve in the i-th speed section, including, The length of the i-th speed segment satisfies: The entry position of the i-th speed section satisfies: s start，i = s end，i -s i (2-10) The guard curve in the i-th speed segment is: In the formula, x is the length of the distance s start，i s start，i ≤ x ≤ s i v start，i is the initial speed of the i-th speed section; a i is the comprehensive deceleration of the i-th speed section; Step S206, judging whether the entrance position of the i-th speed section exceeds the train starting position s start , if yes, taking the train speed v0 as the speed upper limit of the i-th speed section, re-executing steps S203-S205; if no, judging whether the i-th speed section is the last speed section, if yes, ending the algorithm and taking the current speed limit as the protection curve from the train starting position s start to the entrance position s start，i of the i-th speed section; if no, starting to calculate the next speed section, letting i=i+1, v end，i =v start，i-1 , s end，i =s start，i-1 , and re-executing steps S202-S205.

Images