A sub-health diagnosis system and method for rail vehicle door system

Abstract

The present invention discloses a sub-health diagnosis system and method for a rail vehicle door system. The diagnosis method comprises the following steps: first, obtaining the operating state curve of the door system when opening and closing the door under different sub-health states, and performing data preprocessing; then, extracting sensitive features for distinguishing different sub-health states according to the operating state curve, and forming an optimized feature set; combining the support vector machine kernel function with the highest recognition accuracy for each sub-health state to form a multi-core support vector machine for performing sub-health diagnosis of the door system; optimizing the multi-core support vector machine to obtain a sub-health diagnosis model for the door system; finally, extracting features from the door system data to be detected according to the optimized feature set, and using the door system sub-health diagnosis model for diagnosis. The method can determine the sub-health type of a door system that has been identified as being in a sub-health state, and the diagnosis system and method have the advantages of high recognition accuracy and easy expansion.

Description

Technical Field

[0001] The invention belongs to the field of urban rail transit, and in particular relates to a sub-health diagnosis system and method for a rail vehicle door system. Background Art

[0002] With the development of the rail transit industry, the safety and reliability of urban rail door operation have received extensive attention. The high frequency of use of the door system, the complex vehicle operating environment and unknown human factors have caused the number of door failures to account for more than 30% of the total number of vehicle component failures. Therefore, improving the quality and reliability of the door system operation is of great significance to the efficient and safe operation of urban rail trains.

[0003] Sub-health of vehicle door system refers to an intermediate state of the door system. In this state, the system function can still be realized, but its state has deviated from the standard state to a certain extent during the realization process. Further development of this state will lead to one or more types of failures. Conducting sub-health state research can detect abnormalities before the failure actually occurs, effectively prevent the failure from occurring, and maintain the safety of system operation.

[0004] To diagnose sub-health status, it is necessary to accurately classify the sub-health data after obtaining it, so as to determine the specific sub-health type and cause, and then complete the repair of the system. However, the working conditions of the door system are complex, the external uncontrollable influence is large, and the system has many parts and a complex structure. The causes of sub-health are intricate, and the generated data sets are multi-source and heterogeneous. Therefore, after discovering that the door state begins to deviate from the normal state, a diagnostic method is needed to accurately identify the type of sub-health state to meet the needs of repair and maintenance. At present, the local kernel function has local characteristics, strong learning ability but weak generalization ability, and may fall into the local optimal value, and may be difficult to adapt if the data changes. The global kernel function has global characteristics, strong generalization ability but weak learning ability, and it is relatively more difficult to learn the optimal value, and it is also difficult to meet the requirements of simplicity and ease of use. Summary of the invention

[0005] Purpose of the invention: The first purpose of the present invention is to provide a sub-health diagnosis system for a rail vehicle door system. The second purpose of the present invention is to provide a sub-health diagnosis method for a rail vehicle door system.

[0006] Technical solution: A method for diagnosing sub-health of a rail vehicle door system according to the present invention comprises the following steps:

[0007] (1) Obtain the operating status curves of the door system when opening and closing the door under different sub-health conditions, and perform data preprocessing;

[0008] (2) Extract sensitive features for distinguishing different sub-health states based on the operating status curve to form an optimized feature set;

[0009] (3) selecting a support vector machine kernel function according to the recognition effect of each sub-health state, combining the selected kernel functions to form a multi-core support vector machine for performing sub-health diagnosis of the door system; and optimizing the multi-core support vector machine to obtain a door system sub-health diagnosis model;

[0010] (4) After extracting features from the door system data to be tested according to the optimized feature set, the door system sub-health diagnosis model is used for diagnosis.

[0011] Preferably, in step (1), the door system opening and closing process is driven by a motor, and the obtained door system operation status curve includes the stroke, speed, and current status curves generated during the door opening and closing processes.

[0012] Preferably, in step (2), the feature types in the operating status curve are obtained, an original feature set is constructed, the original feature data of different sub-health sample data are combined into a training matrix, and the optimized feature set for distinguishing different sub-health states is obtained after screening using the ReliefF algorithm.

[0013] Preferably, the steps of constructing the original feature set are as follows: according to the operating stage of the door system and the position of the door leaf, the door system's door opening and closing operating status curve is divided into several regions, and the regional features in each region of the operating status curve are extracted; the global features of the door system's door opening and closing operating status curve are extracted; the regional features and the global features in each region of the operating status curve are combined to form the original feature set.

[0014] Preferably, support vector machines with different kernel functions are used to identify the same sub-health state, and the optimal kernel function with the highest recognition accuracy for the sub-health state is obtained. The optimal kernel function corresponding to each sub-health state is assigned weights and then linearly combined to obtain a multi-core support vector machine for sub-health diagnosis of the door system.

[0015] Preferably, the support vector machine kernel function for identifying the same sub-health state includes a linear kernel function, a polynomial kernel function, a Gaussian kernel function, an ANOVA kernel function, a Sigmoid kernel function, a quadratic rational kernel function, a multivariate quadratic kernel function and an inverse multivariate quadratic kernel function.

[0016] Preferably, the particle swarm algorithm is used to iterate the parameters and weights in the constructed multi-core support vector machine until the multi-core support vector machine meets the set accuracy or completes the set number of iteration steps, thereby obtaining the door system sub-health diagnosis model.

[0017] Furthermore, when data preprocessing is performed on the door system operation status curve, the method includes operations such as bad point elimination and data alignment.

[0018] Furthermore, the construction process of the original feature set of the door system operation state curve requires the extraction of detailed and comprehensive feature values ​​in the operation of the door switch. Since the door system movement process has obvious stage characteristics, the operation process of the door system in one cycle is divided into five stages according to the theoretical operating speed, door leaf position and motor movement conditions of the door system, and the regional features of the door system operation curve in each stage and the global features of the entire cycle operation curve are obtained respectively. The global features include the door system opening, time, etc., and the regional features include the maximum value, minimum value, mean, standard deviation, peak-to-peak value, skewness, kurtosis, etc. in each stage;

[0019] Furthermore, since the sub-health state deviates from the healthy state to a smaller extent than the fault state, the original features may contain redundancy and noise that are not conducive to classification. Therefore, the ReliefF algorithm is used to solve such noisy multi-class problems and regression problems. When the ReliefF algorithm is used for feature screening in this method, a training matrix F consisting of several groups of feature data for each type of sub-health state and the corresponding labels are input. The ReliefF algorithm randomly selects a sample point R from the training set, and finds k nearest neighbors H of R from the sub-health class sample set to which the sample point belongs. j (j=1,2,…,k), find k heterogeneous nearest neighbors M from each other heterogeneous sample set j ,According to the distance between R and its neighboring similar samples, and the distance between R and its neighboring heterogeneous samples under each feature, the feature weight is assigned.,Repeat this process and finally get the average weight of each feature.

[0020] The above training matrix Where n represents the total number of samples, m represents the total number of features, and each row of the matrix represents all the features in the original feature set of a data sample. After training the reliefF model, the top j features are selected from the weighted sorting results as the optimized feature set for modeling. The available feature set after dimensionality reduction is obtained

[0021] Furthermore, a sub-health diagnosis model of the door system is constructed based on the multi-core support vector machine. The performance of the support vector machine depends to a large extent on the selection of the kernel function. The multi-core support vector machine selects multiple kernel functions from the basic kernel function set for combined construction. The constructed kernel function can take into account the characteristics of each component kernel function.

[0022] Practice has shown that different kernel functions have different recognition accuracy for the sub-health status of the same door system. For example, the upper slideway inner adjustment, which affects the entire door movement and has a relatively dispersed distribution of feature data, usually works better when using a polynomial kernel or a sigmoid kernel, while the accuracy is highest when using a Gaussian kernel for abnormal sub-health, etc. Therefore, the kernel functions with the highest recognition accuracy for different sub-health states are combined to construct a multi-core SVM with a linear synthetic kernel as the kernel function, which can take into account the advantages of each kernel function, improve the sensitivity and accuracy of identifying multiple types of sub-health, and make the multi-core SVM have better learning ability and better generalization performance than a single kernel, and more accurately classify multiple types of sub-health.

[0023] Furthermore, the particle swarm algorithm is used to iteratively optimize the parameters and weights of each kernel function in the multi-core SVM, so that the sub-health diagnosis model of the door system has higher accuracy. The principle of the particle swarm algorithm is as follows: in the n-dimensional space, the population X = {x1, x2, ..., xm} consists of m particles, where the position of the i-th particle is x i ={x i1 ,x i2 ,…,x in}, speed is v i = {v i1 ,v i2 ,…,v in}, the individual extreme value of the particle is p i ={p i1 , p i2 , ..., p in}, the global extreme value of the entire population is p g ={p g1 , p g2 , ..., p gn After searching for the individual extreme value pbest and the global extreme value gbest, particle x i Will be updated according to the following formula:

[0024] v id (t+1)=ωv id (t)+c1r1(p id (t)-x id (t))+c2r2(p gd (t)-x id (t))

[0025] x id (t+1)=x id (t)+v id (t+1)

[0026] In the formula, t is the current iteration number, x id(t) is the current position of the particle, x id (t+1) is the position of the particle at the next moment, v id (t), v id (t+1) represents the speed of the particle at the current moment and the next moment, ω is the inertia factor, c1 and c2 are acceleration constants, and r1 and r2 are random numbers distributed between [0, 1]. In addition, in order to prevent the particle speed from being too large, a speed upper limit value v is set. max , if v id (t)>v max , take v id (t) = v max , if v id (t)<-v max , take v id (t) = -v max .

[0027] After presetting the hyperparameters of the PSO algorithm, input the initial value of the kernel parameter to be optimized and the training set, perform iterative optimization until the model accuracy meets the set threshold, and then stop the iteration. Finally, output the optimal kernel parameter. Apply the optimized kernel parameter to the kernel function to obtain the optimal multi-core SVM.

[0028] Furthermore, the data to be detected of the door system is data that has been confirmed as sub-health status data but the type of sub-health status has not been identified. The data to be detected is feature screened according to the feature set filtered by the ReliefF algorithm, and the filtered feature data is input into the iteratively optimized multi-core SVM to obtain the sub-health status of the corresponding door system.

[0029] A rail vehicle door system sub-health diagnosis system described in the present invention includes a data receiving device and a sub-health status determination device. The data receiving device is used to receive door system operation status curve data that has been determined to be in a sub-health state, and the sub-health status determination device is used to identify and output the type of sub-health state of the door system.

[0030] Preferably, the sub-health status identification device includes a feature screening module and a sub-health diagnosis module, the feature screening module is used to receive the door system operation status curve data sent by the data receiving device, and extract feature data for distinguishing different sub-health states, the sub-health diagnosis module receives the feature data extracted by the feature screening module, and uses the door system sub-health diagnosis model based on a multi-core support vector machine to identify the type of sub-health state of the door system.

[0031] Preferably, the feature screening module extracts features from the door system operation status curve based on the optimized feature set, and the optimized feature set is obtained by screening the original feature data of different sub-health sample data using the ReliefF algorithm; the multi-core support vector machine in the sub-health diagnosis module is composed of a weighted linear combination of support vector machine kernel functions with the highest recognition accuracy for different sub-health states, and the particle swarm algorithm is used to iterate the parameters and weights in the multi-core support vector machine to obtain a door system sub-health diagnosis model that meets the set requirements.

[0032] Furthermore, the optimized feature set in the feature screening module is in an offline storage state, and the obtained door system operation status curve data is subjected to relevant feature data extraction according to the pre-stored optimized feature set, and the method of obtaining the optimized feature set using the ReliefF algorithm is the same as above.

[0033] Furthermore, the door system sub-health diagnosis model in the sub-health diagnosis module is a pre-stored offline model. The door system sub-health diagnosis model can determine and output the type of sub-health status based on the features extracted by the feature screening module. The construction and optimization method of the door system sub-health diagnosis model is the same as above.

[0034] Beneficial effects: The present invention uses different support vector machine kernel functions to have different recognition accuracy for the same door system sub-health state. When constructing a multi-core support vector machine kernel function, the support vector machine kernel function with the highest recognition accuracy corresponding to each sub-health state is weighted and linearly combined, and a high-precision and practical diagnostic model is obtained after optimization. This diagnostic model takes into account the advantages of excellent learning ability of local kernel functions and better generalization ability of global kernel functions, and can have high recognition accuracy for a variety of door system sub-health states. At the same time, when a new type of door system sub-health state is discovered, it can be quickly iterated, updated and optimized.

[0035] Furthermore, the ReliefF algorithm can be used to address the problem that some operating status curve characteristics are insensitive to sub-health status identification, and sensitive features used to distinguish different sub-health statuses can be screened out from the operating status curve to improve the accuracy of identifying different sub-health statuses; the particle swarm algorithm can be used to iteratively optimize the constructed multi-core support vector machine to obtain the parameters and weights of the kernel function in the multi-core support vector machine, which has the advantage of fast convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the sub-health diagnosis method of the present invention;

[0037] Figure 2 It is a structural principle diagram of a plug sliding door in a specific embodiment of the present invention;

[0038] Figure 3is the classification diagnosis result of the sub-health diagnosis model of the door system in the specific embodiment of the present invention;

[0039] Figure 4 is the single kernel classification diagnosis result of the linear kernel function in the specific embodiment of the present invention;

[0040] Figure 5 It is the single kernel classification diagnosis result of the Gaussian kernel function in the specific embodiment of the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0042] Taking the sub-health detection method of the plug sliding door system commonly used in urban rail subway as an example, the main structure of the plug sliding door is as follows Figure 2 As shown, according to its electromechanical structure, it can be roughly divided into five subsystems: electrical control system, drive locking system, load bearing guide system, internal and external operating system and basic components. The main components include: motor 1, door control system 2, connection device 3, front support 4, rear support 5, nut locking device 6, lock in place switch 7, middle support 8, door in place switch 9, screw rod 10, long guide column 11, end unlocking device 12, internal operating device 13, external operating device 14, left door leaf 15, right door leaf 16. The door control system is used to control the motor to drive the left door leaf and the right door leaf to move synchronously towards each other. One action includes door opening and closing. At the same time, the door control system can also collect the three door state curves of travel, speed and current generated during the door opening and closing process.

[0043] Common sub-health types of plug door systems include: V-shaped dimension abnormality, centering dimension abnormality, buffer head wear, upper slideway inward adjustment, upper slideway outward adjustment and lower gear pin longitudinal interference, etc. The three door status curves of stroke, speed and current generated by the drive motor during the door opening and closing process of the door system in the above sub-health state are obtained respectively, and the operation status curves are subjected to operations such as bad point elimination and data alignment to complete data preprocessing.

[0044] For each group of sub-health types, several groups of door system operation data are obtained, and detailed and comprehensive feature extraction is performed on each group of door system operation data according to the original feature set. In this embodiment, the feature types in the original feature set include global features and regional features. Since the door system movement process has obvious stage characteristics, the operation process of one cycle of the door system is divided into five stages according to the theoretical operating speed of the door system, the position of the door leaf and the movement conditions of the motor, and the regional features of the operation curve of each stage of the door system and the global features of the operation curve of the entire cycle are obtained respectively. Global features include door system opening, time, etc., and regional features include maximum value, minimum value, mean, standard deviation, peak-to-peak value, skewness, kurtosis, etc. in each stage.

[0045] The input is a training matrix composed of multiple types of sub-health feature data and corresponding labels. The ReliefF algorithm is used to extract sensitive features for distinguishing different sub-health states. The training matrix includes all known types of sub-health state data, and each type of sub-health state includes multiple groups of data samples. The specific training matrix is Where n represents the total number of samples, m represents the total number of features, and each row of the matrix represents all the features obtained by a set of data samples based on the original feature set.

[0046] ReliefF randomly selects a sample point R from the training set and finds k nearest neighbors H of R from the sub-health class sample set to which the sample point belongs. j (j=1,2,…,k), find k heterogeneous nearest neighbors M from each other heterogeneous sample set j , according to the distance between R and the similar samples of the nearest neighbors, and the distance between R and the heterogeneous samples of the nearest neighbors under each feature, the feature weight is assigned. Repeat this process and finally get the average weight of each feature. After training with the reliefF algorithm, the top 20 features are selected from the weighted sorting results as the optimized feature set for modeling. That is, the 20 features in the optimized feature set can complete the distinction between different types of sub-health states.

[0047] After the optimized feature set is successfully extracted, an intelligent algorithm is used to construct a sub-health diagnosis model for the door system. In this embodiment, the sub-health diagnosis model for the door system is constructed based on a multi-core support vector machine. The support vector machine kernel function includes a linear kernel function, a polynomial kernel function, a Gaussian kernel function, an ANOVA kernel function, a Sigmoid kernel function, a quadratic rational kernel function, a multivariate quadratic kernel function, and an inverse multivariate quadratic kernel function.

[0048] The selection of kernel functions in the process of constructing a multi-core support vector machine needs to have a good recognition effect on different sub-health states. Using the multi-core linear combination synthesis method, the weight factor is introduced, and the performance of each kernel function on various sub-health states is comprehensively considered. For each sub-health state, the kernel function with the highest recognition accuracy is selected, and after combination, a multi-core support vector machine with a weighted sum kernel as the kernel function is constructed:

[0049]

[0050] Among them, K mix (X,Y) is the kernel function of the multi-core support vector machine, β j is the weight of each single core, β j ≥0 and It is the single-core kernel function that makes up the multi-core.

[0051] After testing, it was found that the sub-health type, such as the upper slide inner adjustment, which affects the entire door movement and has a relatively dispersed distribution of characteristic data, has a higher recognition accuracy when using the polynomial kernel or the sigmoid kernel, while the accuracy of the abnormal sub-health is highest when using the Gaussian kernel. A variety of kernel functions are used to identify each sub-health state, and the kernel function with the highest recognition accuracy corresponding to each sub-health state is selected. Since the same kernel function has the highest recognition accuracy for different sub-health states, in this embodiment, the kernel functions finally selected include linear kernel functions, Gaussian kernel functions, polynomial kernel functions and Sigmoid kernel functions. The specific expressions of the above four kernel functions are as follows: Linear kernel function: K(X,Y)=(X,Y); Polynomial kernel function: K poly (X,Y)=(γX·Y+c) d ; Gaussian kernel function: K Gauss (X,Y)=exp(-γ‖XY‖ 2 ); Sigmoid kernel function: K(X,Y)=tanh(v(X,Y)-c); The kernel function of the multi-core support vector machine is a weighted linear combination of the above four kernel functions.

[0052] In this embodiment, in order to make the door system sub-health diagnosis model have higher recognition accuracy, the particle swarm algorithm is used to optimize the multi-core support vector machine K mix The parameters and weights in (X,Y) are iteratively optimized. The parameters to be optimized include the weight factor β j As well as the kernel parameters γ, c, d of the polynomial kernel and Gaussian kernel, the kernel parameter γ of the Gaussian kernel, and the kernel parameters ν, c of the Sigmoid kernel. The particle encoding thus formed is (γ, c, d, v, β1, β2, β3, β4). Input the label L and the optimized feature set f. Set the threshold T to ensure that the optimization process of the door system sub-health diagnosis model reaches the set accuracy or the number of iterations reaches the set value to stop, and obtain the door system sub-health diagnosis model with the best classification effect.

[0053] When using the optimized door system sub-health diagnosis model, input the door system operation status curve data of a known sub-health state but unknown specific sub-health state type. After data preprocessing, extract 20 sensitive feature data from the door system operation status curve that can distinguish different sub-health states based on the optimized feature set. Input the 20 extracted sensitive feature data into the optimized door system sub-health diagnosis model, and obtain the current sub-health state of the door system through classification.

[0054] In this embodiment, a five-fold cross-validation algorithm is used to calculate the global recognition accuracy of the sub-health diagnosis model of the door system. The results are as follows: Figure 3As shown in Table 1, the global accuracy rate is 99.24%. The recognition accuracy rates for V-shaped anomalies, inner adjustment of the upper slideway, lateral interference of the lower gear pin, centering anomalies, buffer head wear, outer adjustment of the upper slideway, and longitudinal interference of the lower gear pin are shown in Table 1 below:

[0055] Table 1

[0056] Sub-health type Accuracy V-shaped abnormality 100％ Upper slide inner adjustment 100％ Upper slideway external adjustment 97.44％ Lateral interference of lower gear pin 100％ Lower gear pin longitudinal interference 95.49％ Abnormal centering 100％ Buffer head wear 100％

[0057] To show the necessity of feature selection, the global accuracy of the prediction using feature data without feature selection can only reach 28.74%. When the linear kernel function is selected for prediction, the unoptimized traditional single-core SVM is used to predict the data after feature selection. Figure 4 As shown in Table 2, the global accuracy is 94.50%; the recognition accuracy of V-shaped anomaly, upper slideway inward adjustment, lower gear pin lateral interference, centering anomaly, buffer head wear, upper slideway outward adjustment, and lower gear pin longitudinal interference is shown in Table 2:

[0058] Table 2

[0059] Sub-health type Accuracy V-shaped abnormality 100％ Upper slide inner adjustment 91.89％ Upper slideway external adjustment 97.44％ Lateral interference of lower gear pin 100％ Lower gear pin longitudinal interference 80.65％ Abnormal centering 100％ Buffer head wear 88.24％

[0060] When the Gaussian kernel function is selected for prediction, such as Figure 5 As shown in Table 3, the global accuracy is 80.24%; the recognition accuracy of V-shaped anomaly, upper slideway inward adjustment, lower gear pin lateral interference, centering anomaly, buffer head wear, upper slideway outward adjustment, and lower gear pin longitudinal interference is shown in Table 3:

[0061] Table 3

[0062] Sub-health type Accuracy V-shaped abnormality 87.80％ Upper slide inner adjustment 100％ Upper slideway external adjustment 94.87％ Lateral interference of lower gear pin 64.86％ Lower gear pin longitudinal interference 12.90％ Abnormal centering 97.67％ Buffer head wear 85.29％

[0063] Furthermore, when the random forest algorithm is used to predict the types of sub-healthy states, the accuracy rate is 85.35%; when the KNN algorithm is used to predict the types of sub-healthy states, the accuracy rate is 94.13%; when the RBF algorithm is used to predict the types of sub-healthy states, the accuracy rate is 95.12%. It has been verified that the use of ordinary single-core SVM and other existing algorithms, whether it is the global accuracy or the individual diagnostic accuracy of various types of sub-health, is not as good as or even far lower than this solution. In particular, the two sub-healths of longitudinal interference of the lower gear pin and wear of the buffer head have a very large proportion of misclassification, but when using this solution, a higher accuracy classification can be obtained.

[0064] In summary, the algorithm can use the ReliefF algorithm to extract sensitive features for distinguishing different sub-health states for the common sub-health state types of rail vehicle door systems, and improve the accuracy of identifying different sub-health states. The constructed multi-core support vector machine contains the kernel function with the highest recognition accuracy for each sub-health state, taking into account excellent learning ability and better generalization ability. After weight distribution, it has the data mapping ability of multiple kernel functions, and can accurately distinguish and identify different sub-health states with small differences. Through the PSO algorithm, the optimal kernel parameters and weighted parameters are intelligently found, avoiding the problem that the values ​​of methods such as artificial experience setting or grid search deviate from the true optimal solution, which has the advantages of simple operation and fast convergence speed. Furthermore, this method has strong scalability. If new sub-health types appear in the future, it only needs to add new types of training data to the original model, and there is still a certain expectation that the identification can be completed. If the model needs to be expanded, the training data of the new sub-health type can be added, and the kernel parameters can be re-tuned using PSO again, or the performance of various single kernels on the recognition accuracy of the new sub-health type can be comprehensively analyzed to modify the kernel functions involved in the fusion.

[0065] This rail vehicle door system sub-health diagnosis method and system can deeply explore the differences between multiple types of sub-health conditions and intelligently adjust the parameters of the model to establish a diagnosis strategy that is easy to apply and has a higher accuracy than conventional current methods. It is beneficial to improve the operational reliability of the door system, reduce the fault diagnosis pressure of the mainline system facing massive data, and reduce operating costs.

Claims

1. A method for diagnosing sub-health of a rail vehicle door system, characterized in that: The method comprises the following steps: (1) Obtain the operating status curves of the door system when opening and closing the door under different sub-health conditions, and perform data preprocessing; (2) Extract sensitive features for distinguishing different sub-health states based on the operating status curve to form an optimized feature set; (3) selecting a support vector machine kernel function according to the recognition effect of each sub-health state, combining the selected kernel functions to form a multi-core support vector machine for performing sub-health diagnosis of the door system; and optimizing the multi-core support vector machine to obtain a door system sub-health diagnosis model; (4) After extracting features from the door system data to be tested according to the optimized feature set, the door system sub-health diagnosis model is used for diagnosis.

2. The rail vehicle door system sub-health diagnosis method according to claim 1, characterized in that: In the step (1), the door system opening and closing process is driven by a motor, and the door system operation status curve obtained includes the stroke, speed, and current status curves generated during the door opening and closing processes.

3. The rail vehicle door system sub-health diagnosis method according to claim 2, characterized in that: In the step (2), the feature types in the operating status curve are obtained, the original feature set is constructed, the original feature data of different sub-health sample data are combined into a training matrix, and the optimized feature set for distinguishing different sub-health states is obtained after screening using the ReliefF algorithm.

4. The rail vehicle door system sub-health diagnosis method according to claim 3, characterized in that: The steps of constructing the original feature set are as follows: according to the operation stage of the door system and the position of the door leaf, the door system switch operation status curve is divided into several regions, and the regional features in each region of the operation status curve are extracted; the global features of the door system switch operation status curve are extracted; the regional features and global features in each region of the operation status curve are combined to form the original feature set.

5. The rail vehicle door system sub-health diagnosis method according to claim 1, characterized in that: Support vector machines with different kernel functions are used to identify the same sub-health state, and the optimal kernel function with the highest recognition accuracy for this sub-health state is obtained. The optimal kernel function corresponding to each sub-health state is assigned weights and linearly combined to obtain a multi-core support vector machine for sub-health diagnosis of the door system.

6. The method for diagnosing sub-health of a rail vehicle door system according to claim 5, characterized in that: The support vector machine kernel functions for identifying the same sub-health state include linear kernel function, polynomial kernel function, Gaussian kernel function, ANOVA kernel function, Sigmoid kernel function, quadratic rational kernel function, multivariate quadratic kernel function and inverse multivariate quadratic kernel function.

7. The method for diagnosing sub-health of a rail vehicle door system according to claim 5, characterized in that: The particle swarm algorithm is used to iterate the parameters and weights in the constructed multi-core support vector machine until the multi-core support vector machine meets the set accuracy or completes the set number of iterations, and the sub-health diagnosis model of the door system is obtained.

8. A rail vehicle door system sub-health diagnosis system, characterized by: It includes a data receiving device and a sub-healthy state distinguishing device, wherein the data receiving device is used to receive the operating state curve data of the door system that has been judged to be in a sub-healthy state, and the sub-healthy state distinguishing device is used to identify and output the type of the sub-healthy state of the door system; The sub-health status discrimination device includes a feature screening module and a sub-health diagnosis module, wherein the feature screening module is used to receive the door system operation status curve data sent by the data receiving device, and extract feature data for distinguishing different sub-health states, and the sub-health diagnosis module receives the feature data extracted by the feature screening module, and uses a door system sub-health diagnosis model based on a multi-core support vector machine to identify the type of the door system sub-health state; The feature screening module extracts features from the door system operation status curve based on the optimized feature set, and the optimized feature set is obtained by screening the original feature data of different sub-health sample data using the ReliefF algorithm; the multi-core support vector machine in the sub-health diagnosis module is composed of a weighted linear combination of the support vector machine kernel function with the highest recognition accuracy for different sub-health states, and the particle swarm algorithm is used to iterate the parameters and weights in the multi-core support vector machine to obtain a door system sub-health diagnosis model that meets the set requirements.

Images