Training method of snakelike instability detection model and monitoring method of snakelike instability

Abstract

The invention relates to the technical field of railway vehicle operation monitoring, in particular to a training method of a snakelike instability detection model and a snakelike instability monitoring method. The training method comprises the steps of obtaining track data of a wheel-rail contact point when a target vehicle travels in a snaking stable state and track data of the wheel-rail contact point when the target vehicle travels in a snaking unstable state, and extracting artificial features of the track data in the snaking stable state to construct a steady state recognition training set, and extracting artificial features of the trajectory data in the snaking instability state to construct an instability identification training set, and training by using the steady state identification training set and the instability identification training set to obtain a snaking instability detection model. The track data of the wheel-rail contact point can more directly reflect the interaction condition between the wheel and the track during snakelike instability, so that training of the snakelike instability detection model based on the track data of the wheel-rail contact point has higher accuracy in predicting the about-to-enter instability state, and safe operation of a train is guaranteed.

Description

Technical Field

[0001] The present disclosure relates to the technical field of rail vehicle operation monitoring, and in particular to a training method for a snaking instability detection model and a snaking instability monitoring method. Background Art

[0002] The wheel tread of a railway freight car has a certain taper, which causes the self-excited motion of the wheelset, generally manifested as strong anti-phase lateral displacement of the front and rear bogies and strong shaking of the car body, which is called snake instability. There is a critical instability speed for the wheel-rail system of a freight car, and snake instability will occur when the critical speed is exceeded. Snake instability will cause a significant increase in the wear of the wheel tread and the track, and even cause a freight car derailment accident.

[0003] Currently, the monitoring methods for hunting instability can be divided into two categories: the first category is the monitoring standards formulated by various countries. For example, UIC518 points out that if the lateral acceleration of the side frame of a truck reaches 8-10m / s2 for more than 6 consecutive times after a 10Hz low-pass filter, it can be identified as hunting instability; the second category is the algorithm based on the lateral acceleration signal, which mainly uses machine learning, deep learning, signal processing and other technologies to collect or simulate acceleration signals and train the model so that it can automatically identify the acceleration signals during hunting instability.

[0004] The first type of technology mentioned above is an empirical conclusion obtained from a large number of experiments, and it is often impossible to identify when a truck experiences a small degree of snaking instability. For the second type of technology, the acceleration signal is used as the original input for instability identification, and the acceleration signal is used to reflect the vibration of the vehicle during operation.

[0005] However, relying solely on acceleration signals may not be enough to fully describe the specific impact of snaking instability on truck operation, because the same acceleration signal may correspond to snaking motions of different amplitudes. For example, under the same acceleration signal, there may be 0.1mm snaking and 10mm snaking, and the impact of these two situations on truck operation is different. Therefore, snaking instability assessment based on acceleration signals is difficult to accurately reflect the state of vehicle snaking. Summary of the invention

[0006] The present disclosure provides a training method for a snaking instability detection model and a snaking instability monitoring method to solve the problem that snaking instability assessment based on acceleration signals is difficult to accurately reflect the snaking state of a vehicle.

[0007] In a first aspect, the present disclosure provides a training method for a snake instability detection model, comprising:

[0008] Acquire the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking stable state, and the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking unstable state, wherein the track data of the wheel-rail contact points at least include the track data of the wheel-rail lateral contact points;

[0009] Extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking stable state, and constructing a steady-state identification training set based on the artificial features; extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking unstable state, and constructing an unstable identification training set based on the artificial features, wherein the artificial features are features that reflect the distribution characteristics and fluctuation conditions of the trajectory data;

[0010] The steady-state identification training set and the instability identification training set are used to train an initial detection model to obtain a meandering instability detection model.

[0011] In one embodiment, the step of obtaining the track data of the wheel-rail contact point when the target vehicle is traveling in a snaking stable state, and the track data of the wheel-rail contact point when the target vehicle is traveling in a snaking unstable state comprises:

[0012] Constructing a simulation model of the target vehicle traveling on a running track;

[0013] The trajectory data of the wheel-rail contact point in the meandering stable state and the trajectory data of the wheel-rail contact point in the meandering unstable state under the simulation model are obtained.

[0014] In one embodiment, the step of constructing a simulation model of the target vehicle traveling on a running track includes:

[0015] Constructing a simulation model of the target vehicle and the running track, wherein the running track includes a track segment with disturbance and a track segment without disturbance;

[0016] A running route of the target vehicle is constructed, and the running route is used to guide the vehicle to travel through the track section with disturbance and then enter the track section without disturbance.

[0017] In one embodiment, the step of obtaining the trajectory data of the wheel-rail contact point in the snaking stable state and the trajectory data of the wheel-rail contact point in the snaking unstable state under the simulation model comprises:

[0018] Controlling the vehicle to travel along the running route in sequence according to different preset speed values, and obtaining the wheel set lateral displacement of the vehicle after entering the non-disturbance track section corresponding to each speed value, wherein the wheel set lateral displacement is the lateral displacement of the wheel set relative to the center line of the running track when the wheel set is running on the running track;

[0019] Comparing the convergence of the wheelset lateral displacement corresponding to each speed value, determining the serpentine instability critical speed value of the vehicle;

[0020] Controlling the vehicle to travel along the running route at a first speed value lower than the snaking instability critical speed value, and acquiring trajectory data of the wheel-rail contact point in a snaking stable state;

[0021] The vehicle is controlled to travel along the running line at a second speed value higher than the snaking instability critical speed value, and trajectory data of the wheel-rail contact point in a snaking instability state is obtained.

[0022] In one embodiment, the artificial feature includes at least one of the following features:

[0023] Mean, variance, extreme value, skewness, kurtosis, RMS, median, interquartile range, mean absolute error, median absolute difference, mean of differences, median of differences, absolute mean of differences, absolute median of differences, sum of absolute differences, and area under the curve.

[0024] In one embodiment, the steps of extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, constructing a steady-state identification training set according to the artificial features, and extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking unstable state, and constructing an instability identification training set according to the artificial features include:

[0025] extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, performing feature conversion processing on the artificial features to obtain first processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state and the first processing features to form a steady-state feature, and constructing a steady-state recognition training set according to the steady-state features;

[0026] Extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state, performing feature conversion processing on the artificial features to obtain second processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state with the second processing features to form instability features, and constructing an instability identification training set based on the instability features, wherein the feature conversion processing includes performing exponential transformation and / or logarithmic transformation on the artificial features.

[0027] In one embodiment, the step of constructing a steady-state recognition training set according to the steady-state features comprises:

[0028] The importance of each steady-state feature is calculated by using a preset feature screening model, the steady-state features that meet the preset importance rule requirements are screened out, the screened steady-state features are obtained, and the steady-state recognition training set is constructed according to the screened steady-state features;

[0029] The step of constructing an instability identification training set according to the instability features comprises:

[0030] The importance of each of the instability features is calculated using the preset feature screening model, and the instability features that meet the importance rule requirements are screened to obtain the screened instability features. The instability identification training set is constructed based on the screened instability features.

[0031] In one embodiment, the track data of the wheel-rail contact point also includes track data of the wheel-rail vertical contact point.

[0032] In a second aspect, the present disclosure provides a method for monitoring serpentine instability, comprising:

[0033] Acquire the trajectory data of the wheel-rail contact point when the target vehicle is running;

[0034] The trajectory data of the wheel-rail contact point is input into a snake-like instability detection model trained by the method described in any of the above embodiments to obtain a snake-like instability monitoring result.

[0035] In a third aspect, the present disclosure provides a monitoring device for serpentine instability, wherein a serpentine instability detection model trained by the method described in any of the above embodiments is stored on the detection device.

[0036] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the above aspects when executed by a processor.

[0037] In a fifth aspect, the present disclosure provides a computer program product, including a computer program, which implements the steps of the method described in the above aspects when executed by a processor.

[0038] The present disclosure provides a training method for a serpentine instability detection model and a monitoring method for serpentine instability, which trains instability identification by acquiring the trajectory data of the wheel-rail contact point. The trajectory data of the wheel-rail contact point can more directly reflect the interaction between the wheel and the track, especially when serpentine instability occurs, the lateral displacement and contact state between the wheel and the track will change significantly. In contrast, although acceleration data can also reflect the dynamic characteristics of the vehicle, it may be interfered by various external factors (such as wind speed, uneven road surface, etc.), thereby affecting the accuracy of instability judgment. The present disclosure extracts artificial features (such as distribution characteristics and fluctuations) of the trajectory data of the wheel-rail contact point, thereby more accurately describing the difference between serpentine stability and instability, constructing a more accurate steady-state identification training set and instability identification training set, and improving the recognition accuracy and robustness of the detection model. Since the model is constructed directly based on the root cause of serpentine instability (i.e., the lateral interaction between the wheel and the track), the model has a higher accuracy in predicting that the vehicle is about to enter an unstable state, which helps to take measures in advance to prevent the occurrence of instability, thereby ensuring the safe operation of the train. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings:

[0040] Figure 1 A schematic diagram of a flow chart of a method for training a snaking instability detection model provided in an embodiment of the present disclosure;

[0041] Figure 2 A schematic diagram showing a comparison between the lateral displacement of a wheel set of a truck in a stable snaking state and a lateral displacement of a wheel set of a truck in an unstable snaking state provided by an embodiment of the present disclosure;

[0042] Figure 3 A schematic diagram of a training process of a snaking instability detection algorithm provided in an embodiment of the present disclosure;

[0043] Figure 4 A schematic diagram of the reasoning process of a snaking instability detection model provided in an embodiment of the present disclosure;

[0044] Figure 5 A training device for a snake instability detection model is provided in an embodiment of the present disclosure.

[0045] In the drawings, the same reference numerals are used for the same components, and the drawings are not drawn to scale. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, and to fully understand and implement how the present disclosure applies technical means to solve technical problems and achieve the corresponding technical effects, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. The embodiments of the present disclosure and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0047] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0048] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0049] Embodiment 1

[0050] Figure 1 A flowchart of a method for training a snake instability detection model provided by an embodiment of the present disclosure is shown in FIG. Figure 1 As shown, a training method for a snake instability detection model includes:

[0051] Step 110, obtaining the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking stable state, and the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking unstable state, wherein the track data of the wheel-rail contact points at least include the track data of the wheel-rail lateral contact points;

[0052] Step 120, extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking stable state, constructing a steady-state identification training set based on the artificial features, extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking unstable state, constructing an unstable identification training set based on the artificial features, wherein the artificial features are features that reflect the distribution characteristics and fluctuation conditions of the trajectory data;

[0053] Step 130: train an initial detection model using the steady-state identification training set and the instability identification training set to obtain a meandering instability detection model.

[0054] Snake motion is a lateral vibration phenomenon unique to railway vehicles. When a vehicle runs along a straight track at a certain speed, the wheels will show a lateral swing similar to snake due to the geometric relationship between the wheels and the running track and the interaction between the wheels. Swinging within a reasonable range (i.e., snake-like stability) is allowed, but when the swing amplitude is too large or the frequency is abnormal, it may cause the vehicle to become unstable, i.e., snake-like instability occurs, causing the vehicle to be unable to maintain stable driving and even derail.

[0055] The track data of the wheel-rail contact point refers to the position change information of the contact point between the wheel and the running track during driving. The relationship between snaking motion and vehicle driving safety is fundamentally determined by the relative position of the wheel-rail contact point. Therefore, identifying snaking instability through the wheel-rail contact track can more accurately reflect the snaking status of the vehicle.

[0056] The track data of the wheel-rail transverse contact point refers to the position change information of the contact point between the wheel and the running track in a direction perpendicular to the train's travel direction during the train's operation. In this embodiment, the track data of the wheel-rail transverse contact point may include at least the following information:

[0057] 1. Contact point position: The specific lateral position of the contact point between the wheel and the running track at different time points. This position data can reflect the lateral deviation of the wheel during driving;

[0058] 2. Track shape: The track shape formed by connecting a series of contact point position data can reveal the lateral motion trajectory of the wheel, such as whether there is unstable phenomenon such as serpentine motion;

[0059] 3. Change rate: The rate of change of the contact point's position in the lateral direction, i.e., the lateral velocity or acceleration, can reflect the speed and stability of the wheel's lateral movement;

[0060] 4. Timestamp: The time point corresponding to the location data of each contact point is used to determine the time sequence and interval of the data, and then analyze the dynamic change process of the trajectory data.

[0061] The track data of the wheel-rail lateral contact point can be obtained by specific measuring equipment, such as displacement sensors and high-definition cameras installed on the train, to monitor the running posture of the wheelset in real time and collect data. In some embodiments, the track data of the wheel-rail contact point can also be obtained by building a simulation model of the target vehicle and the running track to simulate the actual driving state of the target vehicle.

[0062] Artificial features refer to features that are extracted by artificially analyzing and processing the original wheel-rail contact point trajectory data and that can reflect the distribution characteristics and fluctuations of the trajectory data. In one embodiment, the artificial features include at least one of the following features:

[0063] Mean, variance, extreme value, skewness, kurtosis, root mean square, median, interquartile range, mean absolute error, absolute median difference, difference mean, difference median, difference absolute mean, difference absolute median, sum of difference absolute values ​​and area under the curve. Through the combination of multiple artificial features, the changing trend or fluctuation amplitude of trajectory data can be fully described, so that the steady-state training set constructed based on the artificial features in the serpentine stable state can more accurately reflect the driving characteristics of the target vehicle in the serpentine stable state, and the unstable training set constructed based on the artificial features in the serpentine unstable state can more accurately reflect the driving characteristics of the target vehicle in the serpentine unstable state, thereby improving the accuracy and reliability of subsequent analysis or model prediction.

[0064] The initial detection model can be established using deep learning algorithms (such as convolutional neural networks, recurrent neural networks, etc.) or gradient boosting algorithms (XGBoost algorithms or LightGBM algorithms, etc.). By inputting the steady-state identification training set and the instability identification training set into the initial detection model, the parameters and structure inside the model are continuously adjusted so that the model can accurately distinguish between the two states. The model obtained after training can receive the trajectory data of the wheel-rail contact point in real time, and quickly and accurately determine whether the vehicle is in a snaking instability state, providing important protection for driving safety.

[0065] In summary, the present disclosure provides a training method for a serpentine instability detection model, which obtains the trajectory data of the wheel-rail contact point for instability identification training. The trajectory data of the wheel-rail contact point can more directly reflect the interaction between the wheel and the track, especially when serpentine instability occurs, the lateral displacement and contact state between the wheel and the track will change significantly. In contrast, although the acceleration data can also reflect the dynamic characteristics of the vehicle, it may be interfered by various external factors (such as wind speed, uneven road surface, etc.), thereby affecting the accuracy of instability judgment. The present disclosure extracts artificial features (such as distribution characteristics and fluctuations) of the trajectory data of the wheel-rail contact point, thereby more accurately describing the difference between serpentine stability and instability, constructing a more accurate steady-state identification training set and instability identification training set, and improving the recognition accuracy and robustness of the detection model. Since the model is constructed directly based on the root cause of serpentine instability (i.e., the lateral interaction between the wheel and the track), the model has a higher accuracy in predicting that the vehicle is about to enter an unstable state, which helps to take measures in advance to prevent the occurrence of instability, thereby ensuring the safe operation of the train.

[0066] Embodiment 2

[0067] On the basis of the above embodiment, the step of obtaining the trajectory data of the wheel-rail contact point when the target vehicle is traveling in a snaking stable state and the trajectory data of the wheel-rail contact point when the target vehicle is traveling in a snaking unstable state includes: constructing a simulation model of the target vehicle traveling on a running track; obtaining the trajectory data of the wheel-rail contact point in the snaking stable state and the trajectory data of the wheel-rail contact point in the snaking unstable state under the simulation model.

[0068] In this embodiment, the trajectory data of the wheel-rail contact point is obtained by constructing a simulation model of the target vehicle traveling on the running track. The simulation model can flexibly adapt to various scenarios such as different vehicle types, running speeds, and track conditions, so that sufficient trajectory data of the wheel-rail contact point can be obtained more conveniently and accurately as training sample data, greatly reducing the difficulty of model training.

[0069] In one embodiment, the step of constructing a simulation model of the target vehicle traveling on a running track includes: constructing a simulation model of the target vehicle and the running track, the running track including a track segment with disturbance and a track segment without disturbance; constructing a running route of the target vehicle, the running route being used to guide the vehicle to travel through the track segment with disturbance and then enter the track segment without disturbance.

[0070] In this embodiment, the three-level models of the wheelset, bogie and car body of the target vehicle are successively established using modeling software. After the wheelset modeling is completed, it can be used as a sub-model of the bogie, and after the bogie modeling is completed, it can be used as a sub-model of the car body. The wheelset model includes wheelsets and ballast, and the bogie model includes side frames, axle boxes, bolsters and virtual car bodies. The simulation model of the entire target vehicle may include wheelsets, axle boxes, side frames, bolsters and virtual car bodies. The car body, side frames and wheelsets are all provided with degrees of freedom for activities in different directions; the bolsters can be provided with three degrees of freedom of sinking and floating, shaking their heads and nodding their heads, so as to truly simulate the body swinging state of the target vehicle when it is traveling on the running track. When modeling, only one car can be modeled, and the influence of other adjacent cars on the modeled car can be ignored.

[0071] After that, the wheel tread (the outer cylindrical surface where the wheel contacts the rail) and the running track are modeled. Affected by the wear and damage of the running track, as well as the looseness and compaction of the ballast bed, the geometric state of the line has significant randomness. The combination of multiple factors constitutes the random characteristics of the running track unevenness. The running track unevenness is composed of the superposition of random uneven waves of different wavelengths, phases, and amplitudes. The power spectral density function is usually used to describe the random track unevenness. Therefore, when modeling the running track, the track is divided into a track section with disturbance and a track section without disturbance. The track section with disturbance uses a preset power spectral density function as the input excitation of the track unevenness. The excitation is added to the running line, so that the target vehicle first runs on a section of track with track disturbance, and then enters a smooth track without disturbance, thereby triggering the target vehicle to enter a serpentine instability state.

[0072] In one embodiment, the step of obtaining the trajectory data of the wheel-rail contact point in the snaking stable state and the trajectory data of the wheel-rail contact point in the snaking unstable state under the simulation model comprises:

[0073] The vehicle is controlled to travel along the running route in sequence according to different preset speed values, and the wheelset lateral displacement of the vehicle after entering the undisturbed track section corresponding to each speed value is obtained, wherein the wheelset lateral displacement is the lateral offset of the wheelset relative to the center line of the running track when the wheelset runs on the running track; the convergence of the wheelset lateral displacement corresponding to each speed value is compared to determine the serpentine instability critical speed value of the vehicle; the vehicle is controlled to travel along the running route at a first speed value lower than the serpentine instability critical speed value, and the trajectory data of the wheel-rail contact point in a serpentine stable state is obtained; the vehicle is controlled to travel along the running route at a second speed value higher than the serpentine instability critical speed value, and the trajectory data of the wheel-rail contact point in a serpentine instability state is obtained.

[0074] Specifically, the target vehicle is first run on a track with track disturbance, and then enters a smooth track without disturbance. If the wheelset lateral displacement of the target vehicle on the smooth track shows equal amplitude oscillation or divergence, it means that the speed of the target vehicle has exceeded the nonlinear critical speed. If the wheelset lateral displacement of the target vehicle finally converges, it means that the running speed of the target vehicle is less than the nonlinear critical speed. By continuously adjusting the running speed of the target vehicle for simulation and observing the speed at which the oscillation state of the wheelset lateral displacement converges, it can be determined that this speed is the critical speed of the target vehicle's snaking instability.

[0075] Afterwards, the vehicle is controlled to travel along the running line at a first speed value lower than the critical speed value of the snaking instability, and the trajectory data of the wheel-rail contact point in the snaking stable state is obtained. For example, if the critical speed is 120km / h, the first speed can be 110km / h, 115km / h or 119km / h, etc. The target vehicle is allowed to travel along the running line at different first speeds, and the system obtains the trajectory data of the wheel-rail contact point at different running speeds, so as to obtain the original data for constructing the steady-state identification training set. In the same way, the vehicle is controlled to travel along the running line at a second speed value higher than the critical speed value of the snaking instability, and the trajectory data of the wheel-rail contact point in the snaking instability state is obtained. Among them, the second speed can be 121km / h, 125km / h or 130km / h, etc. The target vehicle is controlled to travel along the running line at different second speeds, and the system obtains the trajectory data of the wheel-rail contact point at different running speeds, so as to obtain the original data for constructing the instability identification training set.

[0076] Embodiment 3

[0077] On the basis of the above embodiment, the steps of extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, constructing a steady-state identification training set according to the artificial features, and extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state, and constructing an instability identification training set according to the artificial features include: extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, performing feature conversion processing on the artificial features to obtain first processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state and the first processing features to form a steady-state feature, and constructing a steady-state identification training set according to the steady-state features; extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state, performing feature conversion processing on the artificial features to obtain second processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state and the second processing features to form an instability feature, and constructing an instability identification training set according to the instability feature, wherein the feature conversion processing includes exponential transformation and / or logarithmic transformation of the artificial features.

[0078] In this embodiment, after the trajectory data of the wheel-rail contact point is obtained, the sliding window method can be used to segment it into a time series, and the artificial features in the trajectory time series can be extracted. After that, the artificial features are subjected to feature conversion processing. In machine learning and data analysis, feature conversion processing is a common data preprocessing step that can improve the learning ability of the model, improve prediction accuracy, or better reveal hidden information in the data. In this embodiment, the method of feature conversion processing can be exponential and / or logarithmic. Transforming the artificial features can change the scale and distribution of the data, help improve the performance of the model, and make the data easier to analyze and learn.

[0079] Specifically, 16 artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state are extracted. The 16 artificial features are subjected to exponential processing and logarithmic processing respectively to obtain the first processing features, which include the 16 artificial features after exponential processing and the 16 artificial features after logarithmic processing. The first processing features are combined with the original 16 features to obtain 48 artificial features. A steady-state recognition training set is constructed based on the combined 48 artificial features. Similarly, the 16 artificial features of the trajectory data of the wheel-rail contact point in the snaking unstable state are subjected to exponential processing and logarithmic processing to obtain the second processing features. The instability recognition training set is jointly constructed based on the second processing features and the 16 artificial features of the trajectory data of the wheel-rail contact point in the original snaking unstable state.

[0080] By taking logarithms and exponentials of artificial features to perform nonlinear transformations, nonlinearity can be introduced to help the model learn more complex patterns. Logarithmic transformation can reduce noise in the data because logarithmic transformation changes slowly in areas with smaller values ​​and changes faster in areas with larger values, thereby smoothing the data to a certain extent, helping to reduce or eliminate data skew, reducing the scale differences of features, and making the data closer to a normal distribution, thereby improving the training effect of the model, making the output of the model easier to interpret, and making the model perform better on different data distributions, thereby improving the generalization ability of the model.

[0081] Embodiment 4

[0082] On the basis of the above embodiment, the step of constructing a steady-state identification training set according to the steady-state features includes: using a preset feature screening model to calculate the importance of each of the steady-state features, screening out the steady-state features that meet the preset importance rule requirements, obtaining the screened steady-state features, and constructing the steady-state identification training set according to the screened steady-state features; the step of constructing an instability identification training set according to the instability features includes: using the preset feature screening model to calculate the importance of each of the instability features, screening out the instability features that meet the importance rule requirements, obtaining the screened instability features, and constructing the instability identification training set according to the screened instability features.

[0083] Generally, important features can more accurately reflect the inherent laws and patterns of the data, thus helping the model to make more accurate predictions. Therefore, in order to improve the training efficiency of the snake instability detection model, the importance of artificial features was screened. By screening out features that contribute little or no to the model's prediction results, the complexity of the model can be reduced, overfitting can be avoided, and the generalization ability and accuracy of the model can be improved.

[0084] In this embodiment, the preset feature screening model can be a tree model, such as a random forest model, which uses the random forest average impurity reduction to screen the features and find the average reduction degree of the error for each feature. The feature importance of all artificial features is calculated and sorted; it can also be a linear model (such as linear regression, logistic regression), and the importance of the feature can be evaluated by the absolute value of the model coefficient. The larger the absolute value of the coefficient, the greater the contribution of the feature to the model prediction result, and therefore the more important it is. The specific feature screening model is selected according to actual needs, and is not specifically limited here.

[0085] Embodiment 5

[0086] Based on the above embodiment, the track data of the wheel-rail contact point also includes track data of the wheel-rail vertical contact point.

[0087] In this embodiment, the track data of the vertical wheel-rail contact point refers to the moving path or coordinate sequence of the contact point in the direction perpendicular to the plane of the track during the contact between the wheel and the running track. In order to more accurately reflect the complex lateral and vertical movements of the wheel during serpentine instability, the track data of the vertical wheel-rail contact point and the track data of the lateral wheel-rail contact point are used to construct the steady-state identification training set and the instability identification training set, which can provide more comprehensive and richer information for the training of the model, improve the accuracy of identifying serpentine instability, and more accurately evaluate the impact of serpentine instability on the train.

[0088] Embodiment 6

[0089] On the basis of the above embodiments, this embodiment provides an application example. This application example is applied to heavy-duty railway freight cars, and the implementation content can be divided into two parts: the first part is the modeling and dynamic simulation of heavy-duty railway freight cars, and the second part is the development and training of the snaking instability algorithm. In the first part, the model of the railway freight car is first established in the Simpack software, and the main parameters such as its wheel tread and rail profile are set. Then, the critical speed of snaking instability of this type of freight car in the current state is determined through the dynamic simulation of the dichotomy method; in the second part, the critical speed of snaking instability is first used as the judgment basis, and a snaking instability detection data set based on the wheel-rail contact trajectory is produced. For the two major variables of the lateral contact trajectory and the vertical contact trajectory in the data set, artificial features are extracted and screened, and the screened artificial features are sent to the established machine learning model to obtain the final output results.

[0090] 1. Modeling and dynamic simulation of heavy-duty railway freight cars

[0091] The three-level models of the wheelset, K6 bogie and car body of the C80 railway freight car are established successively using Simpack software. After the wheelset modeling is completed, it can be used as a sub-model of the bogie, and after the bogie modeling is completed, it can be used as a sub-model of the car body. The wheelset model includes wheelset and ballast, the K6 bogie model includes side frames, axle boxes, bolsters and virtual car bodies, and the whole car model includes the car body. Only one freight car is modeled, and the influence of other adjacent freight cars on the single freight car is ignored. The entire C80 railway freight car includes four wheelsets, eight axle boxes, four side frames, two bolsters, two virtual car bodies and one car body. The car body, side frames and wheelsets all have six degrees of freedom: longitudinal, lateral, sinking and floating, rolling, shaking head and nodding. The bolster has three degrees of freedom: sinking and floating, shaking head and nodding. Therefore, the freight car model has a total of 66 degrees of freedom.

[0092] When using Simpack for multi-body dynamics modeling, its main modeling elements include: bodies, which can be flexible or rigid bodies with mass and moment of inertia; joints, which are used to connect objects or connect to reference coordinate systems; constraints, which are used to limit the degrees of freedom of objects in different directions; force elements, including springs and dampers, which are used to simulate the internal interactions of the system; input functions, which are used to define the input multidimensional data sets; tracks, which are used to define the three-dimensional routes in space that the vehicle can follow; rails, which include information such as rail profiles; rail-wheel pairs, which define the relevant settings for wheel and track contact, including force elements and markers; wheelsets, which define the quasi-linearization settings for wheel-rail contact; and rail-wheel contacts, which define how to calculate the tangential force and tangential torque of the wheel-rail contact patch.

[0093] The wheel tread uses LM type tread, and the running track uses 75kg / m type rail. Affected by the wear and damage of the rails, loosening of the ballast bed, compaction and other factors, the geometric state of the line has significant randomness, and the combination of multiple factors constitutes the random characteristics of track unevenness. Track unevenness is composed of the superposition of random uneven waves of different wavelengths, phases, and amplitudes. The statistical function power spectral density function is usually used to describe the random track unevenness, that is, the track spectrum. At present, many countries such as the United States, Germany, and the United Kingdom have published track spectra. In this embodiment, the track spectrum published by the United States is used, as follows:

[0094] Let S(Ω) represent the track irregularity power spectrum density (cm 2 / (rad / m)), Ω represents the spatial frequency of track irregularity (rad / m), A v , A a Represents the roughness coefficient (cm 2 rad / m), Ω c ,Ω s represents the cutoff frequency (rad / m), k represents the safety factor, and its value is 0.25.

[0095] The power spectrum density of high and low unevenness is:

[0096]

[0097] The power spectrum density of directional irregularity is:

[0098]

[0099] The power spectrum density of horizontal and gauge irregularities is:

[0100]

[0101] In the formula, the orbital level is divided into six levels. Table 1 below shows the parameter values ​​of each orbital level in the United States.

[0102]

[0103] Table 1 US orbit spectrum parameters

[0104] In this embodiment, the American five-level spectrum similar to the design speed and load of my country's heavy-haul railway is selected as the input excitation of track irregularity, and a curved line and a straight line are established respectively, and the track excitation is added to the line.

[0105] The truck is first run on a track with track disturbance, and then enters a smooth track without disturbance. If the lateral displacement of the truck's wheelset on the smooth track shows equal amplitude oscillation or divergence, it means that the speed of the truck has exceeded the nonlinear critical speed. If the lateral displacement of the truck's wheelset finally converges, it means that the running speed of the truck is less than the nonlinear critical speed.

[0106] like Figure 2 The figure shows the comparison of the lateral displacement of the wheelset of the truck in the stable state of snaking and the lateral displacement of the wheelset of the truck in the unstable state of snaking. Through simulation, the lateral displacement of the wheelset converges when the truck runs at a speed of 139km / h, while the lateral displacement of the wheelset oscillates with equal amplitude when the running speed is 140km / h. Therefore, the critical speed of the snaking instability of the truck is 140km / h.

[0107] 2. Development and Training of Snake Instability Algorithm

[0108] The state monitoring of lateral motion stability is the identification of snaking instability. For a certain wheel-rail system, the vehicle will experience snaking instability after reaching a certain speed, which will affect safe operation. Traditional snaking instability identification mainly relies on the vibration acceleration signal of the truck. For example, UIC518 stipulates that after the lateral acceleration of the frame above the axle box is low-pass filtered at 10Hz, the acceleration peak reaches or exceeds 8-10m / s2 for more than 6 consecutive times, which is judged as snaking instability. However, this method is an empirical criterion based on a large number of experiments and lacks theoretical support. The critical speed of snaking instability of the truck is 140km / h, but it is difficult to separate 139km / h and 140km / h by traditional acceleration judgment criteria. This embodiment proposes a state monitoring algorithm for the lateral motion stability of trucks to solve this problem. The trajectory parameters are extracted with artificial features and then input into the model to determine whether snaking instability occurs, which is used to replace the empirical judgment based on acceleration, that is, directly judging whether snaking instability occurs by inputting the wheel-rail contact trajectory, which is used to replace the empirical judgment based on acceleration.

[0109] like Figure 3As shown in the figure, the training process of the entire snake instability detection algorithm includes feature extraction, feature screening, state classification and other steps.

[0110] Through Simpack, simulations were carried out at speeds above and below the critical speed of hunting instability to obtain the wheel-rail lateral contact point trajectory and the wheel-rail vertical contact point trajectory. The sliding window method was used to segment them into time series, and the artificial features in the trajectory time series were extracted and sent to the machine learning algorithm to identify hunting instability, thereby completing the monitoring of the lateral motion stability of the truck.

[0111] Common artificial features of time series include mean, variance, extreme value, skewness, kurtosis, root mean square, median, interquartile range, mean absolute error, absolute median difference, difference mean, difference median, difference absolute mean, difference absolute median, sum of difference absolute values, and area under the curve. As shown in Tables 2 and 3, artificial features are divided according to statistical domain and time domain. Assume that there are N data in a time series, and the i-th data is x i .

[0112]

[0113] Table 2 Statistical domain artificial features

[0114]

[0115]

[0116] Table 3 Temporal artificial features

[0117] After that, artificial features were extracted from the trajectory data of the wheel-rail lateral contact point and the wheel-rail vertical contact point, and a total of 32 features were obtained; the exponential and logarithmic values ​​of the artificial features were taken, and a total of 96 artificial features were obtained by adding the original features. In order to determine which artificial features are more helpful for the classification of this algorithm, the random forest average impurity reduction was used to screen the features, that is, the average reduction degree of each feature on the error was calculated, and the feature importance of all artificial features was calculated and ranked.

[0118] Specifically, artificial feature extraction is performed on the trajectory data of the wheel-rail lateral contact points and the wheel-rail vertical contact points in the serpentine stable state, and the random forest algorithm is used to screen out the screened steady-state features with higher scores, and the screened steady-state features are used to construct a steady-state recognition training set. Artificial feature extraction is performed on the trajectory data of the wheel-rail lateral contact points and the wheel-rail vertical contact points in the serpentine unstable state, and the random forest algorithm is used to screen out the screened instability features with higher scores, and the screened instability features are used to construct an instability recognition training set. The serpentine instability detection model is trained using the steady-state recognition training set and the instability recognition training set.

[0119] In this embodiment, the LightGBM (Light Gradient Boosting Machine) algorithm is used to establish a snake instability detection model. LightGBM is a machine learning algorithm based on a gradient boosting framework, which is used to solve classification and regression problems. In the LightGBM algorithm, the value range of each artificial feature is divided into multiple buckets of equal width, and the samples in each bucket are counted. Then, for each artificial feature, LightGBM selects some representative buckets and uses the indexes of these buckets as new features to replace the original features. By using a Histogram-based algorithm, LightGBM can effectively reduce the number of features and the amount of computation, thereby improving training efficiency and accuracy.

[0120] In addition, the core modules of the LightGBM algorithm include the histogram-based algorithm, the gradient-based one-side sampling technology (GOSS) and the exclusive feature bundling technology (EFB). The traditional random sampling method may lead to an uneven distribution of the training set, thus affecting the performance of the model. When processing an unbalanced data set, the GOSS technology can select the sampled samples according to the gradient information of the data, oversample the samples with larger gradients, and undersample the samples with smaller gradients, so that the distribution of the training set is more balanced. By using the GOSS algorithm, LightGBM can better handle unbalanced data sets, thereby improving the performance and accuracy of the model. The exclusive feature bundling technology is a method for feature selection and processing. Its core idea is to bundle multiple related features together to form a new feature, thereby reducing the number of features and the amount of calculation. It can also automatically select the most representative features, further reducing the workload of manual feature selection, and helping the LightGBM algorithm to better handle high-dimensional data and large-scale data sets, while improving training efficiency and accuracy.

[0121] like Figure 4 , which is a schematic diagram of the reasoning process of the finally trained serpentine instability detection model. In this embodiment, a serpentine instability monitoring method is also provided, including: obtaining the trajectory data of the wheel-rail contact point when the target vehicle is running; inputting the trajectory data of the wheel-rail contact point into the serpentine instability detection model trained by the method described in any of the above embodiments, and obtaining the serpentine instability monitoring result.

[0122] In practical applications, the displacement sensors, high-definition cameras and other equipment installed on the target vehicle can be used to monitor the running posture of the wheelset in real time and collect data to obtain the trajectory data of the wheel-rail contact point when the target vehicle is running. The acquired wheel-rail contact point trajectory data is input into the snaking instability detection model, and the model automatically extracts artificial features and identifies whether it is a snaking instability state based on the extracted artificial features.

[0123] Embodiment 7

[0124] Based on the above embodiments, Figure 5 As shown, this embodiment provides a training device for a snake instability detection model, including:

[0125] The trajectory data acquisition module 510 is used to acquire the trajectory data of the wheel-rail contact point when the target vehicle is traveling in a snaking stable state, and the trajectory data of the wheel-rail contact point when the target vehicle is traveling in a snaking unstable state, wherein the trajectory data of the wheel-rail contact point at least includes the trajectory data of the wheel-rail lateral contact point;

[0126] The training set construction module 520 is used to extract artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, and construct a steady-state identification training set according to the artificial features, and to extract artificial features of the trajectory data of the wheel-rail contact point in the snaking unstable state, and construct an instability identification training set according to the artificial features, wherein the artificial features are features that reflect the distribution characteristics and fluctuation conditions of the trajectory data;

[0127] The detection model training module 530 is used to train an initial detection model using the steady-state recognition training set and the instability recognition training set to obtain a meandering instability detection model.

[0128] In one embodiment, the trajectory data acquisition module includes:

[0129] A modeling unit, used to construct a simulation model of the target vehicle traveling on the running track;

[0130] The data acquisition unit is used to acquire the trajectory data of the wheel-rail contact point in the stable meandering state and the trajectory data of the wheel-rail contact point in the unstable meandering state under the simulation model.

[0131] In one embodiment, the modeling unit comprises:

[0132] A first construction subunit is used to construct a simulation model of the target vehicle and the running track, wherein the running track includes a track segment with disturbance and a track segment without disturbance;

[0133] The second construction subunit is used to construct a running route of the target vehicle, and the running route is used to guide the vehicle to travel through the track section with disturbance and then enter the track section without disturbance.

[0134] In one embodiment, the data acquisition unit includes:

[0135] A control subunit is used to control the vehicle to travel along the running route in sequence according to different preset speed values, and obtain the wheel set lateral displacement of the vehicle after entering the non-disturbance track section corresponding to each speed value, wherein the wheel set lateral displacement is the lateral displacement of the wheel set relative to the center line of the running track when the wheel set is running on the running track;

[0136] A comparison subunit, used for comparing the convergence of the wheelset lateral displacement corresponding to each speed value, and determining the serpentine instability critical speed value of the vehicle;

[0137] A first acquisition subunit is used to control the vehicle to travel along the running route at a first speed value lower than the snaking instability critical speed value, and to acquire trajectory data of the wheel-rail contact point in a snaking stable state;

[0138] The second acquisition subunit is used to control the vehicle to travel along the running line at a second speed value higher than the snaking instability critical speed value, and to acquire trajectory data of the wheel-rail contact point in the snaking instability state.

[0139] In one embodiment, the training set construction module includes:

[0140] a steady-state training set construction unit, configured to extract artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, perform feature conversion processing on the artificial features to obtain first processing features, form steady-state features according to the combination of the artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state and the first processing features, and construct a steady-state recognition training set according to the steady-state features;

[0141] an instability training set construction unit, for extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state, performing feature conversion processing on the artificial features to obtain second processing features, forming instability features according to the combination of the artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state and the second processing features, and constructing an instability identification training set according to the instability features, wherein the feature conversion processing includes performing exponential transformation and / or logarithmic transformation on the artificial features.

[0142] In one embodiment, the steady-state training set construction unit includes:

[0143] A steady-state feature screening subunit is used to calculate the importance of each steady-state feature using a preset feature screening model, screen out the steady-state features that meet the preset importance rule requirements, obtain the screened steady-state features, and construct the steady-state recognition training set according to the screened steady-state features;

[0144] The instability training set construction unit comprises:

[0145] The instability feature screening subunit is used to calculate the importance of each of the instability features using the preset feature screening model, screen out the instability features that meet the importance rule requirements, obtain the screened instability features, and construct the instability identification training set based on the screened instability features.

[0146] The specific definition of the training device for the snaking instability detection model can be found in the definition of the training method for the snaking instability detection model mentioned above, which will not be repeated here. Each unit in the training device for the snaking instability detection model mentioned above can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above units can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above units.

[0147] Embodiment 8

[0148] On the basis of the above embodiments, this embodiment provides a monitoring device for snake-like instability, and the detection device stores a snake-like instability detection model trained by the method described in any of the above embodiments.

[0149] In some implementations of this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0150] In some implementations of this embodiment, a computer program product is provided, including a computer program, which implements the steps of the method described in the above embodiment when executed by a processor.

[0151] The processor may include, but is not limited to, one or more processors or microprocessors, etc. Each processor may be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components to execute the methods in the above embodiments.

[0152] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Computer-readable storage media may include, but are not limited to, for example, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (such as hard disks, floppy disks, solid-state drives, removable disks, Blu-ray disks, etc.).

[0153] The computer-readable storage medium may also store at least one computer executable program, such as a computer-readable instruction. The computer-readable storage medium includes, but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The computer-readable storage medium may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device runs the computer-readable instructions stored on the computer-readable storage medium, the various methods described above may be performed.

[0154] In addition, the serpentine instability monitoring device may also include (but not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (eg, keyboard, mouse, speaker, etc.).

[0155] The processor may communicate with external devices via an I / O bus via a wired or wireless network.

[0156] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product / computer program product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0157] In the embodiments provided in the present disclosure, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0158] It should be noted that in the present disclosure, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element limited by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0159] Although the embodiments disclosed in the present disclosure are as above, the above contents are only embodiments adopted for facilitating the understanding of the present disclosure and are not intended to limit the present disclosure. Any technician in the technical field to which the present disclosure belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present disclosure, but the scope of patent protection of the present disclosure shall still be subject to the scope defined in the attached claims.

Claims

1. A training method for a snake instability detection model, characterized in that: include: Acquire the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking stable state, and the track data of the wheel-rail contact points when the target vehicle is traveling in a snaking unstable state, wherein the track data of the wheel-rail contact points at least include the track data of the wheel-rail lateral contact points; Extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking stable state, and constructing a steady-state identification training set based on the artificial features; extracting artificial features of the trajectory data of the wheel-rail contact point in a snaking unstable state, and constructing an unstable identification training set based on the artificial features, wherein the artificial features are features that reflect the distribution characteristics and fluctuation conditions of the trajectory data; The steady-state identification training set and the instability identification training set are used to train an initial detection model to obtain a meandering instability detection model.

2. The method according to claim 1, characterized in that The step of obtaining the track data of the wheel-rail contact point when the target vehicle is traveling in a snaking stable state, and the track data of the wheel-rail contact point when the target vehicle is traveling in a snaking unstable state comprises: Constructing a simulation model of the target vehicle traveling on a running track; The trajectory data of the wheel-rail contact point in the meandering stable state and the trajectory data of the wheel-rail contact point in the meandering unstable state under the simulation model are obtained.

3. The method according to claim 2, characterized in that The step of constructing a simulation model of the target vehicle traveling on a running track comprises: Constructing a simulation model of the target vehicle and the running track, wherein the running track includes a track segment with disturbance and a track segment without disturbance; A running route of the target vehicle is constructed, and the running route is used to guide the vehicle to travel through the track section with disturbance and then enter the track section without disturbance.

4. The method according to claim 3, characterized in that The step of obtaining the track data of the wheel-rail contact point in the stable meandering state and the track data of the wheel-rail contact point in the unstable meandering state under the simulation model comprises: Controlling the vehicle to travel along the running route in sequence according to different preset speed values, and obtaining the wheel set lateral displacement of the vehicle after entering the non-disturbance track section corresponding to each speed value, wherein the wheel set lateral displacement is the lateral displacement of the wheel set relative to the center line of the running track when the wheel set is running on the running track; Comparing the convergence of the wheelset lateral displacement corresponding to each speed value, determining the serpentine instability critical speed value of the vehicle; Controlling the vehicle to travel along the running route at a first speed value lower than the snaking instability critical speed value, and acquiring trajectory data of the wheel-rail contact point in a snaking stable state; The vehicle is controlled to travel along the running line at a second speed value higher than the snaking instability critical speed value, and trajectory data of the wheel-rail contact point in a snaking instability state is obtained.

5. The method according to claim 1, characterized in that The artificial features include at least one of the following features: Mean, variance, extreme value, skewness, kurtosis, RMS, median, interquartile range, mean absolute error, median absolute difference, mean of differences, median of differences, absolute mean of differences, absolute median of differences, sum of absolute differences, and area under the curve.

6. The method according to claim 1, characterized in that The steps of extracting artificial features of the trajectory data of the wheel-rail contact point in the stable snaking state, constructing a stable state identification training set according to the artificial features, and extracting artificial features of the trajectory data of the wheel-rail contact point in the unstable snaking state, and constructing an unstable identification training set according to the artificial features include: extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state, performing feature conversion processing on the artificial features to obtain first processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking stable state and the first processing features to form a steady-state feature, and constructing a steady-state recognition training set according to the steady-state features; Extracting artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state, performing feature conversion processing on the artificial features to obtain second processing features, combining the artificial features of the trajectory data of the wheel-rail contact point in the snaking instability state with the second processing features to form instability features, and constructing an instability identification training set based on the instability features, wherein the feature conversion processing includes performing exponential transformation and / or logarithmic transformation on the artificial features.

7. The method according to claim 6, characterized in that The step of constructing a steady-state recognition training set according to the steady-state features comprises: The importance of each steady-state feature is calculated by using a preset feature screening model, the steady-state features that meet the preset importance rule requirements are screened out, the screened steady-state features are obtained, and the steady-state recognition training set is constructed according to the screened steady-state features; The step of constructing an instability identification training set according to the instability features comprises: The importance of each of the instability features is calculated using the preset feature screening model, and the instability features that meet the importance rule requirements are screened to obtain the screened instability features. The instability identification training set is constructed based on the screened instability features.

8. The method according to claim 1, characterized in that The track data of the wheel-rail contact point also includes the track data of the wheel-rail vertical contact point.

9. A method for monitoring snake-like instability, characterized in that: include: Acquire the trajectory data of the wheel-rail contact point when the target vehicle is running; The trajectory data of the wheel-rail contact point is input into a snake-like instability detection model trained by the method described in any one of claims 1 to 8 to obtain a snake-like instability monitoring result.

10. A monitoring device for snake-like instability, characterized in that: The detection device stores a hunting instability detection model obtained by training using the method described in any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images