Carbon contact strip abnormal wear early warning method based on multi-working-condition rail transit train operation and maintenance big data

Abstract

The invention discloses a carbon slide plate abnormal wear early warning method based on multi-working-condition rail transit train operation and maintenance big data, and belongs to the technical field of rail transit trains, and the method comprises the following steps: S1, data acquisition, S2, data preprocessing, S3, feature engineering, S4, model construction and training, and S5, real-time early warning. By fully considering various working condition factors such as the running speed, the line condition, the weather condition and the load in the running process of the rail transit train and analyzing and processing the multi-working-condition operation and maintenance big data, the abrasion condition of the carbon contact strip can be more comprehensively reflected, the accuracy and reliability of early warning are improved, and meanwhile, by means of the big data collecting and processing technology, the early warning accuracy is improved. Mass operation and maintenance data are obtained in real time and deep analysis and mining are carried out, abnormal wear characteristics and rules which are difficult to find through traditional manual inspection and single sensor monitoring can be found, and early warning of abnormal wear of the carbon contact strip is achieved.

Description

[0001] The present application relates to the field of rail transit train operation and maintenance technology, more specifically, to a carbon slide plate abnormal wear early warning method based on multi-working-condition rail transit train operation and maintenance big data. BACKGROUND

[0002] In a rail transit system, the carbon slide plate is a key component for the train pantograph to contact the catenary for current collection, and its working state directly affects the safe and stable operation of the train. The carbon slide plate will wear due to friction and other factors during operation, and different operating conditions, such as different line conditions (slope, curve radius, etc.), operating speed, weather conditions (humidity, temperature, wind speed, etc.), and load conditions, will significantly affect the degree of wear of the carbon slide plate.

[0003] The prior art patent document with authorization announcement number CN118643695A discloses a "pantograph carbon slide plate wear monitoring and prediction method, system and medium", including the following steps: S1, obtaining dynamic performance simulation data under different working conditions; S2, obtaining carbon slide plate wear data, forming carbon slide plate wear offline measurement data, real-time contact load between the pantograph and the catenary, and pantograph real-time vibration acceleration data; S3, constructing a dynamic lightweight proxy model; S4, constructing a pantograph carbon slide plate wear prediction virtual simulation model; S5, realizing virtual-real data interaction and information fusion of the pantograph-catenary system physical entity measurement data and the pantograph carbon slide plate wear prediction virtual simulation model; S6, predicting the pantograph carbon slide plate wear and remaining useful life driven by digital twin.

[0004] In addition, the patent document with authorization announcement number CN118898034A discloses a "pantograph carbon slide plate abnormal wear condition identification method and system based on random forest", including the following steps:

[0005] Step 1, collect feature data of pantograph carbon slide plate wear; Step 2, preprocess the feature data, and establish an input matrix and a target vector according to the preprocessed feature data; Step 3, use the input matrix and the target vector as training data to train a random forest regression model, and optimize the parameters of the random forest regression model through cross-validation, to obtain a parameter model of pantograph carbon slide plate abnormal wear; Step 4, based on the feature importance index of the parameter model of pantograph carbon slide plate abnormal wear, identify the key causes of pantograph carbon slide plate abnormal wear;

[0006] Although the prior art realizes real-time calculation and life prediction of wear by constructing a carbon slide plate wear prediction system based on multi-working condition dynamic simulation, combining dynamic modeling and virtual simulation, and using a random forest algorithm to preprocess and model train carbon slide plate wear characteristic data to identify abnormal wear key causes, the prior art fails to comprehensively integrate multi-source heterogeneous real-time operation and maintenance data in the data fusion dimension, making it difficult to reflect the comprehensive influencing factors of carbon slide plate wear. In addition, the feature engineering method is limited to static feature extraction, lacking dynamic optimization and dimensionality reduction processing of multi-working condition combination features, and the early warning architecture design does not form a closed-loop mechanism of'real-time collection-intelligent analysis-dynamic early warning', resulting in high early warning delay and insufficient generalization ability, which cannot meet the timeliness and accuracy requirements of abnormal wear early warning of rail transit operation and maintenance. SUMMARY

[0007] The present application mainly provides a carbon slide plate abnormal wear early warning method based on multi-working condition rail transit train operation and maintenance big data, which can solve the problems raised in the above background art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a carbon slide plate abnormal wear early warning method based on multi-working condition rail transit train operation and maintenance big data, comprising:

[0009] First, the carbon slide plate wear amount, contact pressure, running speed, line slope, weather condition data of multiple trains under different operating conditions are collected, and the running time and train number identification information corresponding to each group of data are recorded in detail to provide comprehensive and accurate raw data support for subsequent analysis and processing. Then, the collected raw data is carefully cleaned and arranged, missing data is filled in through professional methods, abnormal values are removed, and data dimensions are unified to effectively improve data quality and lay a reliable foundation for model training. Then, single-working condition features and multi-working condition combination features related to carbon slide plate abnormal wear are deeply extracted from the preprocessed data, key factors affecting wear are fully mined, and the pertinence and accuracy of early warning are significantly improved. Then, a carbon slide plate abnormal wear early warning model is constructed using machine learning algorithms, model parameters are optimized through a large amount of data training, the model is given intelligent judgment ability for carbon slide plate wear state, and accurate prediction of abnormal wear is realized. Finally, the optimized model is applied to the train operation process, dynamic operation and maintenance data are collected and processed in real time, and when abnormal wear signs are found, an early warning signal is quickly sent to ensure that the staff can timely grasp the equipment state and effectively ensure the safety of train operation.

[0010] Further, the data preprocessing specifically includes interpolation method and filling missing data using machine learning algorithm, and abnormal data is identified and processed through Z-score method and IQR method, and the data is normalized to unify the data range of different dimensions.

[0011] Further, in the feature engineering, the single working condition features include carbon slide plate wear rates under different running speeds and contact pressures under different line slopes; the multi-working condition combination features include the combined effects of running speed and line slope and the combined effects of weather conditions and load, and the correlation analysis and principal component analysis statistical methods are adopted for dimension reduction and high contribution degree features are reserved.

[0012] Further, the machine learning algorithm for constructing the carbon slide plate abnormal wear early warning model is a neural network algorithm, the preprocessed data are divided into a training set and a test set, multi-working condition features are taken as input and carbon slide plate abnormal wear states are taken as output for supervised learning training, and the generalization ability of the optimized model is evaluated through cross validation.

[0013] Further, the real-time early warning specifically refers to real-time collection of train operation and maintenance data, extraction of feature parameters after processing according to the data preprocessing and feature engineering steps, input of the early warning model, timely issuance of an early warning signal to notify the operation and maintenance personnel when the model judges that there is abnormal wear.

[0014] Further, the data collection is realized by real-time collection of wear amounts, contact pressures, vibrations, carbon strip XYZ axis displacement amounts by sensors installed on the carbon slide plate, simultaneous acquisition of train running speeds, line slopes, curve radii, contact network voltages and currents, weather conditions and train load data, and recording of running time, train number and carbon slide plate number identification information.

[0015] Further, the multi-working condition factors include running speed, line condition, weather condition and train load, and through multi-working condition operation and maintenance big data analysis, the carbon slide plate wear condition is comprehensively reflected, and the early warning accuracy is improved.

[0016] The beneficial effects of the carbon slide plate abnormal wear early warning method based on multi-working condition rail transit train operation and maintenance big data are as follows:

[0017] By fully considering various operating factors such as the operating speed, line conditions, weather conditions, and load during the operation of rail transit trains, and analyzing and processing multi-operating operation and maintenance big data, the wear of the carbon slide can be more comprehensively reflected, and the accuracy and reliability of the early warning can be improved. At the same time, with the help of big data collection and processing technology, massive operation and maintenance data can be obtained in real time and in-depth analysis and mining can be carried out. Abnormal wear characteristics and patterns that are difficult to detect through traditional manual inspections and single sensor monitoring can be discovered, and early warning of abnormal wear of the carbon slide can be achieved. In addition, by applying the constructed early warning model to the actual rail transit train operation process, train operation and maintenance data is collected in real time, and input into the model after step-by-step processing, abnormal wear of the carbon slide can be discovered in time, avoiding train failures and accidents caused by abnormal wear, reducing the frequency and cost of manual inspections, and improving the operation and maintenance efficiency and safety of rail transit trains. Finally, the early warning model constructed using machine learning algorithms can be adjusted and optimized according to different types of rail transit trains, operating lines, and operating conditions. It has strong scalability and adaptability, which improves the versatility and practicality of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0019] Fig. 1 This is a schematic flow chart of the steps of a carbon slide abnormal wear early warning method based on multi-operating rail transit train operation and maintenance big data of the present invention;

[0020] Fig. 2 The present invention is a method flow chart of a carbon slide abnormal wear early warning method based on multi-operating-condition rail transit train operation and maintenance big data. DETAILED DESCRIPTION

[0021] In order to make the technical solution of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1

[0023] like Figs. 1-2 As shown, a technical solution is provided: a carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data, comprising the following steps:

[0024] Step 1: Data collection:

[0025] Through the displacement sensor, pressure sensor and vibration sensor installed on the carbon slide plate, the carbon slide plate wear amount, contact pressure, vibration data and carbon strip XYZ axis displacement amount are collected in real time, the line parameters such as train running speed, line slope and curve radius are synchronously acquired, and the environmental data such as contact network voltage, current, environmental temperature and humidity, and wind speed are synchronously acquired, each group of data is uniquely identified by using time stamp and equipment number (train number, carbon slide plate number) (for example, {time stamp, train ID, carbon slide plate ID, wear amount, contact pressure, speed, slope, temperature}), and a multi-dimensional original data set is formed,

[0026] The line slope calculation formula is:

[0027]

[0028] Where Δh is the height difference, and Δs is the horizontal distance.

[0029] Step 2, data preprocessing:

[0030] Then the collected original data is carefully cleaned and arranged, missing data is filled by professional method, abnormal values are removed, and data dimension is unified, which effectively improves the data quality and lays a reliable foundation for model training. Specifically, the interpolation method and machine learning algorithm are used to fill the missing data, and the Z-score method and IQR method are used to identify and process abnormal data. At the same time, the data is normalized to unify the data range of different dimensions.

[0031] The interpolation method is to calculate the interpolation result by using the adjacent time data (x1, x2) and (y1, y2) through the following formula:

[0032]

[0033] For non-linear missing mode, a random forest (machine learning algorithm) regression model is used to predict the missing wear amount or contact pressure value by taking time, speed, slope, etc. as features,

[0034] The abnormal data is processed by calculating the mean μ and standard deviation σ of the data by the Z-score method. For data point x, if:

[0035]

[0036] It is determined as an abnormal value, and the IQR method is used to assist verification. The quartiles Q1 and Q3 are calculated. If IQR = Q3-Q1, the abnormal threshold is Q1-1.5IQR or Q3+1.5IQR. Double verification ensures the accuracy of abnormal value identification,

[0037] In addition, the normalization processing is to use min-max normalization, the formula is:

[0038]

[0039] Where x is the original data, x min and x max are the minimum and maximum values of the feature, respectively, and x ′ is the normalized data. Through this formula, data of different dimensions such as wear (0-50mm) and contact pressure (0-100N) can be uniformly mapped to the [0,1] interval, ensuring that the model's weight distribution for each feature is not affected by the original value range, and improving the stability and accuracy of model training.

[0040] Step 3, feature engineering:

[0041] Then, single-condition features and multi-condition combination features related to carbon slide plate abnormal wear are deeply extracted from the preprocessed data, key factors affecting wear are fully mined, and the pertinence and accuracy of the warning are significantly improved. Specifically, single-condition features include carbon slide plate wear rate at different running speeds and contact pressure at different line slopes; multi-condition combination features include the combined effect of running speed and line slope, and the combined effect of weather conditions and load. Correlation analysis and principal component analysis statistical methods are used for dimension reduction, and high contribution features are retained,

[0042] Where single-condition feature extraction includes the following operations:

[0043] Wear rate calculation: For different running speed intervals, calculate the wear rate per unit distance, the specific formula is:

[0044]

[0045] For example, the average wear rate at 60-80km / h,

[0046] Contact pressure feature: Calculate the contact pressure mean and variance under different line slopes, such as the contact pressure fluctuation coefficient when the slope is >5%, which is represented as:

[0047]

[0048] And the multi-condition combination feature construction is:

[0049] Speed-slope combination: Construct the feature υ×slope to reflect the synergistic effect of high speed and large slope on wear, the principle is the coupling effect of load and speed in the theory of friction and wear,

[0050] Weather-load combination: Calculate the Pearson correlation coefficient of humidity and train load:

[0051]

[0052] and the combined features with |r|>0.5 are screened, such as the coupling feature of high humidity (>80%) and heavy load (load>120% of rated value),

[0053] Finally, the screening and dimensionality reduction need to remove redundant features and compress the data dimension through statistical methods. First, the mutual information of each feature and the abnormal wear label is calculated by correlation analysis, and the key features with mutual information greater than 0.1 are retained. The mutual information calculation formula is:

[0054]

[0055] Where p(x, y) is the joint probability distribution of feature X and label Y, p(x) and p(y) are the marginal probability distributions, and the higher the information value, the stronger the correlation between the feature and the wear state. For the retained features, further dimensionality reduction is performed through principal component analysis (PCA), and the covariance matrix is calculated:

[0056]

[0057] Then, the principal components with cumulative variance contribution rate exceeding 95% are extracted, for example, the first k principal components satisfy:

[0058]

[0059] Where λ i is the eigenvalue of the covariance matrix, which reduces the data dimension while retaining key information and improves the model calculation efficiency.

[0060] Step 4, model construction and training:

[0061] Then, a machine learning algorithm is used to construct a carbon slide abnormal wear early warning model. Through a large amount of data training and optimization of model parameters, the model is given the intelligent judgment ability of the carbon slide wear state, and the precise prediction of abnormal wear is realized. Specifically, the machine learning algorithm for constructing the carbon slide abnormal wear early warning model is a neural network algorithm. The preprocessed data is divided into training set and test set, and supervised learning training is performed with multi-working-condition features as input and carbon slide abnormal wear state as output. The model generalization ability is evaluated and optimized through cross-validation,

[0062] Where the neural network algorithm adopts a three-layer BP neural network, the number of input layer nodes is consistent with the feature dimension output by the feature engineering (such as 20-dimensional features after principal component analysis), the number of hidden layer nodes is set to 2 times the input layer + 1 (41 nodes), the number of output layer nodes is 2 (corresponding to "normal" "abnormal" binary classification), and the activation function is ReLU (linear rectifier function), the formula is:

[0063] ReLU(x)=max(0,x)

[0064] The function can alleviate the gradient vanishing problem and improve the model training efficiency,

[0065] In addition, the training process is closely combined with parameter optimization and generalization ability evaluation, which jointly guarantees the model performance. First, the preprocessed data set is divided into a training set (8000) and a test set (2000) according to an 8:2 ratio. The cross-entropy loss function is used:

[0066]

[0067] The error of the binary classification task is measured. The Adam optimizer (learning rate η = 0.001, weight decay coefficient λ = 0.001) is used for 500 iterations. A 5-fold cross-validation mechanism is introduced to divide the training set into 5 parts for cyclic verification. The average F1 score is calculated:

[0068]

[0069] The generalization ability is evaluated. If the F1 score of the validation set is less than 0.85, the number of hidden layer neurons or the learning rate is dynamically adjusted, and the model is retrained until it reaches optimal performance on the validation set.

[0070] Step 5, real-time warning:

[0071] Finally, the optimized model is applied to the train operation process. Real-time collection and processing of dynamic operation data are performed. When abnormal wear signs are found, a warning signal is quickly issued to ensure that workers can timely grasp the equipment status and effectively ensure train operation safety.

[0072] The above-described embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as limiting the scope of the present patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present patent should be subject to the appended claims.

Claims

1. A carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data, characterized by: First, data on carbon slide wear, contact pressure, operating speed, track gradient, and weather conditions were collected from multiple trains under different operating conditions. The operating time and train number identification information corresponding to each set of data were recorded in detail to provide comprehensive and accurate raw data support for subsequent analysis and processing. The collected raw data was then meticulously cleaned and organized, using professional methods to fill in missing data, remove outliers, and standardize data dimensions, effectively improving data quality and laying a solid foundation for model training. Then, the single working condition characteristics and multi-working condition combination characteristics related to abnormal wear of the carbon slide are deeply extracted from the preprocessed data, and the key factors affecting wear are comprehensively explored, which significantly improves the pertinence and accuracy of the early warning; then, the machine learning algorithm is used to construct an abnormal wear early warning model for the carbon slide, and the model parameters are optimized through large-scale data training, giving the model the ability to intelligently judge the wear status of the carbon slide, and realizing accurate prediction of abnormal wear; finally, the optimized model is actually applied to the train operation process, and dynamic operation and maintenance data is collected and processed in real time. When signs of abnormal wear are found, a warning signal is quickly issued to ensure that the staff can grasp the equipment status in a timely manner and effectively ensure the safety of train operation.

2. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: The data preprocessing specifically includes filling missing data using interpolation and machine learning algorithms, identifying and processing abnormal data through the Z-score method and IQR method, and normalizing the data to unify the data ranges of different dimensions.

3. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: In the feature engineering, single-operating-condition features include the carbon slide wear rate at different operating speeds and the contact pressure at different line slopes; multi-operating-condition combination features include the combined influence of operating speed and line slope, and the combined influence of weather conditions and load. Correlation analysis and principal component analysis statistical methods are used to screen and reduce dimensions and retain features with high contribution.

4. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: The machine learning algorithm for constructing the carbon slide abnormal wear warning model is a neural network algorithm. The preprocessed data is divided into a training set and a test set. Supervised learning training is performed with multi-working condition characteristics as input and the abnormal wear status of the carbon slide as output. The generalization ability of the optimized model is evaluated through cross-validation.

5. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: The real-time warning specifically involves collecting train operation and maintenance data in real time, extracting feature parameters after data preprocessing and feature engineering steps, and inputting them into a warning model. When the model determines that there is abnormal wear, a warning signal is promptly issued to notify the operation and maintenance personnel.

6. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: The data acquisition collects the wear amount, contact pressure, vibration, and carbon bar XYZ axis displacement in real time through sensors installed on the carbon slide, and simultaneously obtains the train running speed, line slope, curve radius, contact network voltage and current, weather conditions and train load data, and records the running time, train number, and carbon slide number identification information.

7. The carbon slide abnormal wear warning method based on multi-operating rail transit train operation and maintenance big data according to claim 1 is characterized by: The multi-operating condition factors include operating speed, line conditions, weather conditions, and train load. Through multi-operating condition operation and maintenance big data analysis, the wear condition of the carbon slide plate can be fully reflected and the accuracy of the early warning can be improved.

Images