CN114462700A - Railway track quality evaluation method and device - Google Patents

Abstract

This paper provides a railway track quality evaluation method and device. The method includes: acquiring detection data of the track geometry of the railway track; importing the detection data into a prediction model to determine the vehicle response of the railway track; The vehicle response is used to determine the track quality of the railway track, and the easily obtained detection data of the track geometry of the railway track can be converted into the vehicle response through the prediction model. The vehicle response can replace the track geometry to predict the track quality, which can solve the problem of The track geometry on the railway track does not exceed the limit, but the problem of invisible diseases that affect the safety and comfort of train operation provides support for scientific evaluation of track quality and guidance for reasonable maintenance.

Description

A kind of railway track quality evaluation method and device

technical field

The invention relates to the technical field of rail transit, can be applied to general-speed trunk lines, and in particular is a method and device for evaluating the quality of a railway track.

Background technique

At present, a scientific track quality evaluation method is an important prerequisite for reflecting the true quality of the track, guiding reasonable maintenance and improving the safety and comfort of railway operation.

In order to evaluate the quality of the track, the railway department uses the track inspection vehicle to periodically test the railway track of the general-speed trunk line. The testing items mainly include the track geometric parameters such as height, track direction, gauge, level, triangle pit, etc., and calculate the track geometric peak value and Mathematical statistics such as segment standard deviation evaluate track quality.

However, there is a complex relationship between track geometry and vehicle dynamic performance. The overrun of track geometry does not necessarily cause greater vehicle vibration. On the contrary, there may be an overrun of vehicle response at certain positions where the track geometry is not overrun. Ordinary-speed trunk lines are usually used to transport passengers. In order to improve the user experience of passengers, railway departments need to focus on the safety and comfort of passengers in the car.

The research results of this paper show that the most directly related to the safety and comfort in the vehicle is the vehicle dynamic performance, or the vehicle response. Therefore, the existing track quality evaluation methods based on track geometric statistics cannot objectively reflect the actual vehicle response, and thus cannot find some invisible track geometric defects that affect the safety and comfort of train operation.

SUMMARY OF THE INVENTION

In view of the above problems in the prior art, the purpose of this paper is to provide a railway track quality evaluation method and device, so as to solve the problem that the existing track quality evaluation methods based on track geometric statistics in the prior art cannot objectively reflect the actual vehicle response, thereby It is impossible to find some invisible problems of track geometry that affect the safety and comfort of train operation.

In order to solve the above technical problems, the specific technical solutions in this paper are as follows:

On the one hand, this paper provides a railway track quality evaluation method, including:

Obtain the detection data of the track geometry of the railway track;

importing the detection data into a prediction model to determine the vehicle response of the railway track;

From the vehicle response, a track mass of the railway track is determined.

As an embodiment of this document, before the detection data is imported into a prediction model and the vehicle response of the railway track is determined, the method includes:

The detection data is subjected to wavelet decomposition to obtain several wavelength components;

Classify several wavelength components according to the wavelength range;

importing several types of the wavelength components into the prediction model;

Among them, the track geometry includes left height, right height, left track, right track, gauge, level and triangle pit.

As an embodiment of this paper, the prediction model is obtained by the following training method:

Selecting a training subset from the training set and inputting it into the prediction model to obtain a predicted value of the vehicle response corresponding to the wavelength component;

determining the mean square error loss of the training subset according to the predicted value of the vehicle response and the actual value of the vehicle response corresponding to the training subset;

According to the regularization coefficient, the model parameters and the mean square error loss, the regularization loss is obtained;

In reverse time series, sequentially calculate the partial derivatives of the regularization loss with respect to the hidden states of the prediction model;

updating the model parameters according to the partial derivative, the hidden state, the model parameters and a preset learning rate of the prediction model;

Repeat the above steps until the preset number of training steps or when the reduction rate of the regularization loss is less than the preset transformation threshold, stop the training of the prediction model.

As an embodiment of this document, after the training is stopped, the method further includes:

Verifying whether the prediction model satisfies the verification condition, and when the prediction model satisfies the verification condition, it is determined that the training of the prediction model is completed;

The verification conditions include:

The prediction of the prediction model is determined according to the comparison images drawn by the training set and the validation set at the regularization loss of each training set, and the comparison images of the vehicle response predicted by the prediction model stopped training in the validation set and the real vehicle response. whether the results are accurate;

If inaccurate, expand the training set and retrain the prediction model.

If it is accurate, it is determined that the training of the prediction model is successful;

As an embodiment of this document, the determining whether the prediction result of the prediction model is accurate further includes:

According to the root mean square error formula and the Hill inequality coefficient formula, evaluate whether the prediction result of the prediction model whose training is stopped is accurate;

The root mean square error formula is

The formula for Hill's inequality coefficient is

分别表示验证集中车辆响应真实值y和预测值

的第i个值，i＝1,2,…,M，M为验证集容量；Among them, yi and

Represents the true value y and the predicted value of the vehicle response in the validation set, respectively

The i-th value of , i=1,2,...,M, where M is the verification set capacity;

If the result obtained by the root mean square error formula satisfies the first error threshold, and the result obtained by the Hill inequality coefficient formula satisfies the second error threshold, it is judged that the prediction result of the prediction model is accurate.

As an embodiment of this document, before selecting a training subset from the training set and inputting it into the prediction model, before obtaining the predicted value of the vehicle response corresponding to the wavelength component, the steps include:

确定所述轨道几何和所述车辆响应中每一维度的规范化数据，得到规范化数据，其中x表示一个维度的所述轨道几何，σ表示该维度下所述轨道几何的标准差，

表示该维度下所述轨道几何的均值；According to the normalized formula

Determine the normalized data of each dimension in the track geometry and the vehicle response to obtain normalized data, where x represents the track geometry in one dimension, σ represents the standard deviation of the track geometry in this dimension,

represents the mean of the orbital geometry in this dimension;

Selecting several sections in the normalized data as the training set and the validation set;

The model parameters and the hidden state are initialized with random numbers to obtain the prediction model.

As an embodiment of this document, the determining the mass of the railway track in each section according to the vehicle response further includes:

According to the vehicle response, calculate the comfort index and safety index of the railway track in each section;

The comfort index and the safety index are evaluated against limit values to determine the railway track quality for each section.

As an embodiment of this document, calculating the comfort index and safety index of the railway track in each section according to the vehicle response further includes:

The vehicle response includes vehicle body vertical acceleration, vehicle body lateral acceleration, left wheel vertical force, right wheel vertical force, left wheel lateral force and right wheel lateral force;

The comfort index includes the vertical acceleration of the vehicle body and the lateral acceleration of the vehicle body;

The safety indicators include derailment coefficient, wheel load reduction rate and wheel-axle lateral force;

Wherein, the derailment coefficient is calculated according to the lateral force of the left wheel, the lateral force of the right wheel, the vertical force of the left wheel and the vertical force of the right wheel;

The wheel weight reduction rate is calculated according to the reduction amount of the vertical force of the wheel-rail relative to the average static wheel weight and the average static wheel weight, wherein the average static wheel weight is the average of the vertical force of the left wheel and the vertical force of the right wheel under static conditions value;

The axle lateral force is calculated from the left wheel lateral force and the right wheel lateral force.

As an embodiment of this document, evaluating the comfort index and the safety index according to the limit value to determine the railway track quality in each section further includes:

The limits include comfort limits and safety limits;

the comfort level limit includes a first limit value and a second limit value;

The safety limit includes a third limit, a fourth limit and a fifth limit;

determining whether the vertical acceleration of the vehicle body exceeds the first limit value, and determining whether the lateral acceleration of the vehicle body exceeds the second limit value, if none of them exceeds, the comfort index of the section meets the standard;

Determine whether the derailment coefficient exceeds the third limit value, determine whether the wheel load reduction ratio exceeds the fourth limit value, determine whether the lateral force of the wheel axle exceeds the fifth limit value, and if none exceeds the , then the safety index of this section meets the standard;

Taking the section where both the comfort index and the safety index meet the standard as the section with good railway track quality;

A section for which at least one of the comfort index and the safety index fails to meet the standard is regarded as a section with poor railway track quality.

On the other hand, a railway track quality evaluation device is also provided, comprising:

an acquisition unit for acquiring the detection data of the track geometry of the railway track;

an importing unit for importing the detection data into a prediction model to determine the vehicle response of the railway track;

a determining unit, configured to determine the track quality of the railway track according to the vehicle response.

By adopting the above technical solution, the easily obtained detection data of the track geometry of the railway track can be converted into the vehicle response through the prediction model. However, the problem of invisible diseases that affect the safety and comfort of train operation provides support for scientific evaluation of track quality and guidance for reasonable maintenance. In order to make the above-mentioned and other objects, features and advantages of this paper more obvious and easy to understand, preferred embodiments are hereinafter described in detail in conjunction with the accompanying drawings.

Description of drawings

In order to more clearly illustrate the embodiments or the technical solutions in the prior art, the following briefly introduces the accompanying drawings that are used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only For some embodiments herein, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative effort.

Fig. 1 shows the overall system diagram of a method for evaluating the quality of a track railway track according to an embodiment of this paper;

FIG. 2 shows a schematic diagram of waveform decomposition of a method for evaluating the quality of a track railway track according to an embodiment of this paper;

FIG. 3 shows a schematic diagram of steps of a method for evaluating the quality of a track railway track according to an embodiment of this document;

FIG. 4 shows a schematic diagram of a method for evaluating the quality of a railroad track according to an embodiment of this document;

Fig. 5(a) and Fig. 5(b) show a vehicle body acceleration comparison diagram according to the embodiment of this paper;

Fig. 6(a)-Fig. 6(d) show a wheel-rail force comparison diagram of the embodiment of this paper;

FIG. 7 shows a schematic diagram of quality determination of a railway track quality evaluation method according to an embodiment of this paper;

FIG. 8 shows a schematic diagram of a railway track quality evaluation device according to an embodiment of this document;

FIG. 9 shows a flowchart of a method for evaluating the quality of a railway track according to an embodiment of this document;

FIG. 10 shows a schematic diagram of the processing flow of a prediction model of a railway track quality evaluation method according to an embodiment of this document;

FIG. 11 shows a schematic diagram of the track geometry of the embodiment of this paper;

FIG. 12 shows a schematic diagram of the vehicle response according to the embodiment of this paper;

FIG. 13 shows a schematic diagram of the computer equipment of the embodiment of this document.

Description of the symbols in the drawings:

101. Track inspection vehicle;

102. database;

103. Computing server;

104. Control terminal;

801. Obtain unit;

802. Import unit;

803. Determine the unit;

1302. Computer equipment;

1304. processor;

1306. memory;

1308. Drive mechanism;

1310. Input/output module;

1312. Input device;

1314. Output device;

1316. Presentation equipment;

1318. Graphical user interface;

1320, network interface;

1322. Communication link;

1324. A communication bus.

Detailed ways

The technical solutions in the embodiments herein will be clearly and completely described below with reference to the accompanying drawings in the embodiments herein. Obviously, the described embodiments are only a part of the embodiments herein, rather than all the embodiments. Based on the embodiments herein, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection herein.

It should be noted that the terms "first", "second" and the like in the description and claims herein and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific sequence or sequence. It is to be understood that data so used may be interchanged under appropriate circumstances such that the embodiments herein described can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having", and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device comprising a series of steps or units is not necessarily limited to those expressly listed Rather, those steps or units may include other steps or units not expressly listed or inherent to these processes, methods, products or devices.

As shown in FIG. 1 , an overall system diagram of a method for evaluating the track quality of a track railway includes a track inspection vehicle 101 , a database 102 , a computing server 103 and a control terminal 104 .

Among them, the track inspection vehicle 101 is equipped with an accelerometer and a force measuring wheel pair corresponding to the railway track. Through the accelerometer and the force measuring wheel pair, on the track to be measured, the track inspection vehicle 101 can detect the geometric dimensions of the entire railway track of several general-speed trunk lines. In this embodiment, the track geometry includes the left height and the right height. , left track, right track, gauge, level and triangular pit, these seven parameters characterize the physical characteristics of the railway track.

In this embodiment, the samples corresponding to the training set of the prediction model come from known railway tracks, and the track inspection vehicle 101 can run on known tracks and record its own vehicle response. It should be noted that different trains, Running on the same railroad track, the resulting vehicle response may be different. The main reason is that the number of vehicles, the load capacity or the type of vehicles are different, and these reasons restrict the vehicle response obtained by the train running on the railway track. Herein, the vehicle response may include vehicle body vertical acceleration, vehicle body lateral acceleration, left wheel vertical force, right wheel vertical force, left wheel lateral force, right wheel lateral force.

The database 102 is used to store the detected data of the track inspection vehicle 101, and can also store the detected data corresponding to the track geometry obtained by the normal running trains except the track inspection vehicle 101. It should be noted that there are Various sensors can correspondingly collect detection data of seven kinds of track geometries. The detection data takes the mileage as a sequence, with the starting point as the point of t=0 and the end point as T, seven types of detection data can be established. The database 102 stores the seven kinds of detection data in a unified manner with a certain track, a certain train, and a certain mileage as an identifier.

As shown in FIG. 2 , a schematic diagram of the decomposition of detection data of a method for evaluating the track quality of a track railway is shown. The computing server 103 can divide the seven types of detection data into three types according to wavelength ranges. The computing server 103 uses the Bior4.4 wavelet basis to perform 7-level discrete wavelet decomposition on the detection data of the track geometry to obtain the high-frequency coefficients d1, d2, d3, d4, d5, d6, d7 and low-frequency coefficients a7 at all levels, and then analyze the d1- d4 is reconstructed to obtain short-wave components, and the corresponding spatial wavelength is less than 8m; d5-d7 is reconstructed to obtain medium-wave components, and the corresponding spatial wavelengths are 8-64m; a7 is reconstructed to obtain long-wave components, and the corresponding spatial wavelength is greater than 64m. The detection data is decomposed and reconstructed by wavelet, and the data of different bands are shown in Figure 2.

In Figure 2, the leftmost side of each rectangular box represents the amplitude of the waveform. It should be noted that the amplitude is calculated by the wavelength and frequency of the detection data. For the convenience of description, this paper uses the amplitude as the detection data. The ordinate of , and the mileage as the abscissa of the detection data.

It should be noted that the data waveforms of three different wavelengths are exemplarily given in this article, and those skilled in the art can also add types according to wavelength classification as needed, for example, they can be divided into 4, 5, 6 and other types The data waveform is not limited in this paper.

The computing server 103 is also provided with a prediction model, and the prediction model can output the vehicle response corresponding to the train according to the detection data of the three wavelengths,

The control terminal 104 is used for according to the instruction of the dispatch center, for example, the operator of a certain train, or the conductor of the train, said that in a certain shift, the operation is uncomfortable or unsafe, then the control terminal 104 can call the database 102. The data of the train is imported into the prediction model, and the prediction model is used to evaluate which sections have poor track quality or poor track quality, so as to remind the railway track operation and maintenance personnel to go to repair.

In the prior art, the track quality is often judged by the track geometry. At the position where the track geometry exceeds the limit, those skilled in the art generally believe that the position needs to be repaired, so the position is regarded as the section where the track quality does not meet the standard, and maintenance personnel are required to carry out In this paper, through objective analysis and the conclusion drawn from a large number of experiments, there is no simple one-to-one correspondence between the track geometry and whether the passengers are comfortable or safe when the vehicle is running. Passengers may also feel comfortable or safe in a position where the track geometry does not exceed the limit, passengers may feel uncomfortable or unsafe.

Therefore, the existing track quality evaluation methods based on track geometric statistics cannot objectively reflect the actual vehicle running state, and thus cannot find some invisible track geometric defects that affect the safety and comfort of train operation.

In order to solve the above problems, the embodiment of this paper provides a method for evaluating the track quality of a railroad railway, which can find some invisible diseases of the track geometry that affect the safety and comfort of train operation. FIG. 3 is a railroad track quality provided by the embodiment of this paper. Schematic diagram of the steps of the evaluation method, the present specification provides the method operation steps as described in the examples or flow charts, but may include more or less operation steps based on routine or non-creative work. The sequence of steps enumerated in the embodiments is only one of the execution sequences of many steps, and does not represent the only execution sequence. When an actual system or device product is executed, the methods shown in the embodiments or the accompanying drawings may be executed sequentially or in parallel. Specifically, as shown in Figure 3, the method may include:

Step 301 , acquiring detection data of the track geometry of the railway track.

Step 302: Import the detection data into a prediction model to determine the vehicle response of the railway track.

Step 303: Determine the track quality of the railway track according to the vehicle response.

By adopting the above technical solution, the easily obtained detection data of the track geometry of the railway track can be converted into the vehicle response through the prediction model. However, the problem of invisible diseases that affect the safety and comfort of train operation provides support for scientific evaluation of track quality and guidance for reasonable maintenance.

As an embodiment of this document, step 301 obtains the waveform corresponding to the track geometry of the railway track, which may include:

The waveform corresponding to the track geometry is obtained through the sensors of the track inspection vehicle running on the railway track. The waveform takes the mileage as the sequence, the train departure point as the point of t=0, and the train end point as T.

As an embodiment of this document, before the waveform is imported into the prediction model in step 302 and the vehicle response of each section of the railway track is determined, the method includes:

The detection data is subjected to wavelet decomposition to obtain several wavelength components.

Several wavelength components are classified by wavelength range.

Several types of the wavelength components are imported into the prediction model.

Among them, the track geometry includes left height, right height, left track, right track, gauge, level and triangle pit.

In this step, considering that the orbit geometry has multiple wavelength components, in order to make the calculation of the subsequent prediction model more accurate, the detection data of the multiple orbit geometries can be decomposed and reconstructed.

Using the Bior4.4 wavelet basis, 7-level discrete wavelet decomposition is performed on the detection data of orbital geometry to obtain high-frequency coefficients d1, d2, d3, d4, d5, d6, d7 and low-frequency coefficients a7.

The short-wave components are obtained by reconstructing d1-d4, and the corresponding spatial wavelength is less than 8m.

Reconstructing d5-d7 to obtain the medium wave component, the corresponding spatial wavelength is 8-64m;

The long-wave component is obtained by reconstructing a7, and the corresponding spatial wavelength is greater than 64m.

In this way, the waveforms of the detection data of several wavelengths can be classified. Compared with the original data corresponding to the seven orbital geometries, the statistical characteristics of the three types of wavelength components are easier to be learned by the prediction model operation. When the prediction model can adapt to the components of the three types of statistical features, the computing efficiency and accuracy can be improved.

It should be noted that the embodiments of this paper illustrate that seven detection data corresponding to seven orbital geometries can be converted into three wavelength components. Similarly, the detection data can also be converted into more wavelength components, which is not limited in this paper. .

As shown in FIG. 4, a schematic diagram of a method for evaluating the track quality of a track railway is shown. As an embodiment of this paper, in step 302, the prediction model can be obtained by the following training method:

Step 401: Select a training subset from the training set and input it into the prediction model to obtain a predicted value of the vehicle response corresponding to the wavelength component.

In this step, on a known railway track, according to the mileage range of the track, the data of 80% of the mileage with continuous mileage is selected as the training data set, the data of 10% of the mileage with continuous mileage is selected as the verification data set, and the continuous mileage of 10% of the mileage data is selected as the verification data set. % of the data is used as the test data set; the mileage ranges of the training set, validation set, and test set do not overlap each other.

B (10-100) subsequences are randomly selected from the training set as a training subset, and the length of each sequence is T.

Assuming that the actual mileage of the railway track corresponding to the training subset is 100m, then the sequence length T=400 (the spatial sampling interval of the track geometry data is 0.25m). Organize the training data into a 3-dimensional tensor format, denoted as {x _B , y _B }, that is, the dimension of x _B is B×T×n, and the dimension of y _B is B×T×m.

Step 402: Determine the mean square error loss of the training subset according to the predicted value of the vehicle response and the actual value of the vehicle response corresponding to the training subset.

Input the three types of wavelength components corresponding to each area on the railway track into the prediction model. At the current time node, the prediction model can be characterized by the model parameter ω and the hidden state h, and input the three types of wavelength components corresponding to each area. In the prediction model, the predicted value of the vehicle response can be obtained, which can be expressed by formula (1):

表示短波成分在预测模型中的输出，

表示中波成分在预测模型中的输出，

表示长波成分在预测模型中的输出。

为车辆响应预测值。in,

represents the output of the shortwave component in the prediction model,

represents the output of the medium wave component in the prediction model,

Represents the output of the longwave components in the prediction model.

Predicted value for vehicle response.

In this paper, according to formula (2), the mean square error loss of the training subset is calculated, and the specific formula is as follows:

In formula (1), B, T and m are the dimensions of the training subset, y _B is the real value of the vehicle response, j, i and k are the serial numbers of the training subset, where the real value of the vehicle response is based on the track inspection vehicle by the The predicted value of the vehicle response is obtained by the prediction model based on the detection data of the track geometry.

Step 403: Obtain the regularization loss according to the regularization coefficient, the model parameters and the mean square error loss.

In this step, in order to prevent overfitting of the regularization term of the prediction model, it is necessary to calculate the regularization loss, according to formula (3):

Loss=loss+λ∑|ω| ² (3)

Wherein, loss is derived from formula (2), ω is the model parameter of the prediction model in this prediction, and λ represents the regularization coefficient. In this embodiment, the regularization coefficient can be understood as a constant.

Step 404: Calculate the partial derivatives of the regularization loss with respect to the hidden state of the prediction model in reverse time series.

In this step, the detection data of the track geometry is described according to the mileage from the initial point to the end point, and in this paper, the mileage can be determined through time series, for example, the speed of the track inspection vehicle can be determined, which can be obtained at a certain time. sequence of miles.

In order to modify the model parameters of the prediction model in the way of gradient backpropagation according to the structural characteristics of the model connected by time, the examples in this paper are as follows:

From T to 0 along the time reverse sequence, calculate the partial derivative of the regularization loss relative to the hidden state h _i of the ith time step in turn, using the following formula (4):

In the formula, c=1, 2, 3, indicating the output of the short-wave component in the prediction model, the output of the medium-wave component in the prediction model, and the output of the long-wave component in the prediction model.

The partial derivative of the regularization loss LOSS relative to the hidden state h _i of the i-th prediction model can be obtained. It can be seen in the formula that the formula also requires the prediction model at the next moment of the current time point corresponding to the hidden state h _{i+ 1.} For example, the current partial derivative calculated by formula (4) is the third time, then the third time can be calculated only after the fourth time is completed. If the partial derivative calculation of the fourth time is not completed, the fourth time must be completed first. Partial derivative calculation of time.

Step 405: Update the model parameters according to the partial derivatives, the hidden states, the model parameters, and a preset learning rate of the prediction model.

In this step, in order to update the prediction model, it is necessary to update the model parameters of the trained prediction model, and the new model parameters can be solved by using formula (5):

In the above formula (5), the model parameter on the right side of the equation is the model parameter of the current prediction model, and the partial derivative of the root mean square error with respect to the model parameter and the learning rate calculated by formula (3) are used to obtain the intermediate variable, The calculated model parameters can be obtained by subtracting the intermediate variables from the model parameters of the current prediction model. In this paper, the learning rate can be constant.

In the above manner, the prediction model can be adjusted, and when the next training subset is input to the prediction model, the model parameters obtained by formula (5) are used as the model parameters of the prediction model.

Step 406: Repeat the above steps until the preset number of training steps or when the reduction rate of the regularization loss is less than the preset transformation threshold, stop the training of the prediction model.

In this step, in order to reduce the amount of training and avoid the waste of a large amount of computing resources, the number of training steps can be set according to the empirical value. In this paper, the number of training steps can be set to 5000, that is, the cycle of steps 401-405 is 5000 After that, the training is terminated.

Alternatively, in order to more accurately judge whether the model has been trained, it can also be judged according to the reduction rate of the regularization loss, that is, it can be judged according to the regularization loss obtained from the two training intervals, and the current regularization loss can be subtracted from the above. The one-time regularization loss is then compared with the current regularization loss, and if the absolute value of the ratio is less than 1%, the training of the prediction model is stopped.

As an embodiment of this document, after the training is stopped, the method further includes:

Verifying whether the prediction model satisfies the verification condition, and when the prediction model satisfies the verification condition, it is determined that the training of the prediction model is completed;

The verification conditions include:

The prediction of the prediction model is determined according to the comparison image drawn by the validation set, the regularization loss of the validation set at each training period, and the comparison image of the vehicle response predicted by the prediction model stopped training in the validation set and the real vehicle response. whether the results are accurate;

If inaccurate, expand the training set and retrain the prediction model.

If it is accurate, it is determined that the training of the prediction model is successful.

In this step, after the training is stopped, in order to prove that the current prediction model can better fit the real results, the trained prediction model needs to be verified.

Whether the training of the prediction model is completed can be determined by setting the verification conditions, the output prediction model can be executed by the method of determining whether the prediction model satisfies the verification conditions, or the model can be retrained by expanding the training set.

When the prediction model satisfies the verification conditions, the prediction model is output as an exemplary illustration of this article. In order to avoid the waste of computing resources, the prediction model can usually be trained once a quarter. When the prediction model training is completed and the verification conditions are passed, the The prediction model is pushed to all railway tracks, or railway tracks of general-speed trunk lines for use.

When the prediction model fails to meet the verification conditions, the track inspection vehicle can be made to obtain the track geometry data of the known track again, and the data length and sequence length can be adjusted to increase the sample dimension of the training set. When importing into the prediction model, the types of wavelengths can be added. For example, the track geometry can be divided into four types, five types, etc. according to the wavelength to increase the training accuracy of the prediction model, or the number of training steps can be increased, for example, to 10,000 times. Or reduce the reduction rate of the regularization loss.

It should be noted that the calculation method of the reduction rate can be 1-(last regularization loss/current regularization loss).

As an embodiment of this document, the step of determining whether the prediction result of the prediction model is accurate according to the image drawn by the validation set and the image drawn by the prediction model that stopped training further includes:

According to the root mean square error formula and the Hill inequality coefficient formula, evaluate whether the prediction result of the prediction model whose training is stopped is accurate;

The root mean square error formula is

The formula for Hill's inequality coefficient is

分别表示验证集车辆响应真实值y和车辆响应预测值

的第i个值，i＝1,2,…,M，M为验证集容量；Among them, yi and

Represents the true value y of the vehicle response in the validation set and the predicted value of the vehicle response

The i-th value of , i=1,2,...,M, where M is the verification set capacity;

If the result obtained by the root mean square error formula satisfies the first error threshold, and the result obtained by the Hill inequality coefficient formula satisfies the second error threshold, it is judged that the prediction result of the prediction model is accurate.

In this step, waveform comparison can be used to verify the accuracy of the prediction model, such as a vehicle body acceleration comparison diagram as shown in Figure 5(a) and Figure 5(b), and Figure 6(a)- Figure 6(d) shows a wheel-rail force comparison diagram. It can be seen that the prediction result of the prediction model can be obtained by comparing the predicted value output by the prediction model and the actual value waveform, and the difference between the two can be compared.

In the embodiment of this paper, in order to describe the dispersion degree of all the data predicted by the model relative to the actual value on the long railway track, the root mean square error formula needs to be used to calculate the root mean square error. If the error is particularly large, the railway track operation and maintenance personnel can be misled, and the result obtained by the root mean square error formula can be used as one of the verification conditions.

In the embodiments herein, in order to evaluate the relative prediction errors of the models and facilitate comparison between different models, a Hill inequality coefficient with an amplitude between 0 and 1 is used.

As an embodiment of this document, in step 401, a training subset is selected from the training set and input into the prediction model, and before obtaining the predicted value of the vehicle response corresponding to the wavelength component, the method includes:

确定所述轨道几何的检测数据和所述车辆响应中每一维度的规范化数据，得到规范化数据，其中，yi和

分别表示验证集中车辆响应真实值y和预测值

的第i个值，i＝1,2,…,M，M为验证集容量；According to the normalized formula

Determine the detection data of the track geometry and the normalized data of each dimension in the vehicle response to obtain normalized data, where yi and

Represents the true value y and the predicted value of the vehicle response in the validation set, respectively

The i-th value of , i=1,2,...,M, where M is the verification set capacity;

Several segments are selected in the normalized data as the training set and the validation set.

The model parameters and the hidden state are initialized with random numbers to obtain the prediction model.

In this step, in order to speed up the convergence speed of the prediction model training process and avoid Z-shaped convergence, it is necessary to normalize the data input before the training model.

In this embodiment, the waveform of the original track geometry may be normalized, or the wavelength components classified into three types may be normalized, which is not limited herein.

For example, if the orbit geometry is normalized, the orbit geometry is divided into seven dimensions according to the type, and the mean and standard deviation of each dimension are calculated respectively.

确定七种维度的轨道几何对应的规范化数据。After the calculation is complete, according to the normalized formula

Normalized data corresponding to orbital geometry in seven dimensions is determined.

The normalized data represents the waveform of the entire railway track. According to the ratio of 8:1:1, the training set, test set and validation set are selected from the normalized data.

When the prediction of the prediction model is performed for the first time, a random number can be taken from the hidden state of the prediction model to obtain the initial prediction model.

As shown in FIG. 7 , a schematic diagram of quality determination of a railway track quality evaluation method, as an embodiment of this paper, in step 303, according to the vehicle response, the railway track quality of each section is determined, further comprising:

Step 701: Calculate the comfort index and safety index of the railway track in each section according to the vehicle response.

Step 702: Evaluate the comfort index and the safety index according to the limit value to determine the railway track quality of each section.

In this step, the quality of the railway track can be judged by the comfort and safety of the train running on the railway track.

It has been proved by a large number of experimental data, as well as the feedback of passengers, that comfort and safety are related to the audience degree of passengers for a certain flight.

Therefore, this paper selects the above comfort index and safety index to evaluate the quality of the railway track in a certain section. That is, the user experience is used as the testing standard for the quality of the railway track.

As an embodiment of this document, the step 701 of calculating the comfort index and safety index of the railway track in each section according to the vehicle response, further includes:

The vehicle response includes vehicle body vertical acceleration, vehicle body lateral acceleration, left wheel vertical force, right wheel vertical force, left wheel lateral force, and right wheel lateral force.

The comfort index includes the vertical acceleration of the vehicle body and the lateral acceleration of the vehicle body.

The safety indicators include derailment coefficient, wheel load reduction rate and wheel-axle lateral force.

Wherein, the derailment coefficient is calculated according to the lateral force of the left wheel, the lateral force of the right wheel, the vertical force of the left wheel and the vertical force of the right wheel.

The wheel weight reduction rate is calculated according to the reduction amount of the vertical force of the wheel-rail relative to the average static wheel weight and the average static wheel weight, wherein the average static wheel weight is the average of the vertical force of the left wheel and the vertical force of the right wheel under static conditions value.

The axle lateral force is calculated from the left wheel lateral force and the right wheel lateral force.

In this step, the vehicle response can be divided into 6 types. In this paper, according to the long-term investigation, the comfort index can be attributed to the vertical acceleration of the vehicle body and the lateral acceleration of the vehicle body, that is, the bumps of the vehicle body will affect the comfort of passengers Therefore, the classification method conforms to the laws of nature.

In this step, the safety indicators can also be attributed to the derailment coefficient, the wheel load reduction rate and the lateral force of the wheel and axle. The above three safety indicators are obtained according to a large number of experiments, that is, according to the accident factors in the accident vehicle The arrangement of weights also conforms to the laws of nature.

To sum up, it is in line with the laws of nature to use the above two indicators to judge the quality of the railway track in a certain section.

Among them, the lateral force of the left wheel and the lateral force of the right wheel can be obtained to obtain the lateral force of the wheel-rail, and the vertical force of the left wheel and the vertical force of the right wheel can be obtained to obtain the vertical force of the wheel-rail. derailment factor.

Calculate the average static wheel weight with the average value of the left wheel vertical force and the right wheel vertical force of the train under static state, subtract the average static wheel weight from the wheel-rail vertical force to obtain the load reduction amount, and divide the load reduction amount by the average static wheel weight. Wheel weight gets the wheel load reduction rate.

The vertical force of the left wheel and the vertical force of the right wheel can be obtained from the technical manual of the train under static condition, and will not be repeated here.

As an embodiment of this document, in step 702, evaluating the comfort index and the safety index according to the limit value to determine the railway track quality in each section, further includes:

The limits include comfort limits and safety limits.

The comfort level limit includes a first limit value and a second limit value.

The safety limits include a third limit, a fourth limit, and a fifth limit.

It is determined whether the vertical acceleration of the vehicle body exceeds the first limit value, and whether the lateral acceleration of the vehicle body exceeds the second limit value is determined.

Determine whether the derailment coefficient exceeds the first limit value, determine whether the wheel load reduction ratio exceeds the second limit value, determine whether the lateral force of the wheel axle exceeds the third limit value, if none of them exceeds the limit. , the safety index of this section meets the standard.

The section in which both the comfort index and the safety index meet the standard is regarded as the section with good railway track quality.

A section for which at least one of the comfort index and the safety index fails to meet the standard is regarded as a section with poor railway track quality.

In this step, the first limit can be set as 0.6m/s ² , the second limit can be set as 0.6m/s ² , the third limit can be set as 0.8, the fourth limit can be set as 0.8, and the fifth limit can be set as 0.8. The value is set as 10+P ₀ /3, where P ₀ is the net axle load, which can be obtained from the technical manual of the train.

As shown in Table 1, the limit values of comfort index and safety index are as follows:

Table 1

As can be seen in Table 1, when the vertical acceleration of the car body is less than 1.0m/s ² and the lateral acceleration of the car body is less than 0.6m/s ² , the quality of the railway track in this section can be considered to be up to the standard. If one is not satisfied, it is considered that the comfort index of this section is not up to standard.

When the derailment coefficient of the car body is less than 1.0, the derailment rate of the wheel weight of the car body is less than 0.8, and the lateral force of the wheel axle is less than 10+P ₀ /3, the quality of the railway track in this section can be considered to be up to the standard. If it is not satisfied, it is considered that the safety index of this section is not up to the standard.

When the comfort index and safety index of a certain section of the railway all meet the standards, the quality of the section is deemed to be qualified and no maintenance is required.

This paper also provides a schematic diagram of a railway track quality evaluation device as shown in Figure 8, including:

The acquiring unit 801 is used for acquiring the detection data of the track geometry of the railway track.

The importing unit 802 is configured to import the detection data into a prediction model to determine the vehicle response of the railway track.

A determination unit 803, configured to determine the track quality of the railway track according to the vehicle response.

Through the above device, it is possible to convert the easily obtained detection data of the track geometry of the railway track into the vehicle response through the prediction model. The vehicle response can replace the track geometry to predict the track quality, and the track geometry on the railway track can be solved. The problem of invisible diseases affecting the safety and comfort of train operation provides support for scientific evaluation of track quality and guidance for reasonable maintenance.

This paper also provides a flow chart of a railway track quality evaluation method as shown in Figure 9, including:

Step 901: Acquire the detection data of the track geometry and perform data decomposition and reconstruction. This step corresponds to the specific text content of step 301 .

Step 902: Input the reconstructed detection data into the prediction model to obtain the vehicle response. In this step, technical inspiration can be obtained according to the schematic diagram of the processing flow of the prediction model of a railway track quality evaluation method as shown in Figure 10. In Figure 10, LSTM1 corresponds to the processing unit of the short-wave components in the prediction model, and LSTM2 corresponds to the processing of the medium-wave components in the prediction model. Unit, LSTM3 corresponds to the processing unit of long-wave components in the prediction model. The processing results obtained by LSTM1, LSTM2 and LSTM3 are summed and weighted with LOSS to obtain the vehicle response.

Step 903: Calculate the comfort index and safety index of each section of the railway track according to the vehicle response. This step is described in detail through the part shown in step 701 .

Step 904: Determine whether the comfort index and safety index of a certain section both meet the limit value, if so, go to step 905, if not, go to step 906. This step is described in detail through the part shown in step 702 .

Step 905 , the section belongs to the quality standard section and does not need maintenance.

Step 906 , the input quality of this section is not up to the standard section, and maintenance is required.

For step 906, in this step, it can be concluded from the track geometry diagram shown in FIG. 11 that at the position marked by the gray box, the vertical acceleration and height of the train are not abnormal, but the passengers are uncomfortable at this position Therefore, through the prediction model of this paper, the schematic diagram of the vehicle response shown in Figure 12 is obtained. It can be seen that at the position corresponding to the gray box in Figure 11, the weight reduction rate of the right wheel and the right height of the vehicle body are abnormal, and It is concluded that the safety index of this section is not up to standard, which is consistent with the feedback of passengers, so the maintenance personnel need to be notified to maintain this section.

As shown in FIG. 13 , for a computer device provided by the embodiments herein, the computer device 1302 may include one or more processors 1304 , such as one or more central processing units (CPUs), each processing unit may implement One or more hardware threads. The computer device 1302 may also include any memory 1306 for storing any kind of information such as code, settings, data, and the like. Without limitation, for example, memory 1306 may include any one or a combination of the following: any type of RAM, any type of ROM, flash memory devices, hard disks, optical disks, and the like. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of computer device 1302. In one instance, when processor 1304 executes the associated instructions stored in any memory or combination of memories, computer device 1302 may perform any operation of the associated instructions. The computer device 1302 also includes one or more drive mechanisms 1308 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, and the like.

Computer device 1302 may also include an input/output module 1310 (I/O) for receiving various inputs (via input device 1312 ) and for providing various outputs (via output device 1314 ). A specific output mechanism may include presentation device 1316 and associated graphical user interface (GUI) 1318. In other embodiments, the input/output module 1310 (I/O), the input device 1312 and the output device 1314 may not be included, and only serve as a computer device in the network. Computer device 1302 may also include one or more network interfaces 1320 for exchanging data with other devices via one or more communication links 1322. One or more communication buses 1324 couple together the components described above.

Communication link 1322 may be implemented in any manner, eg, through a local area network, a wide area network (eg, the Internet), a point-to-point connection, etc., or any combination thereof. Communication links 1322 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

Corresponding to the method in FIG. 2-FIG. 10, the embodiments herein also provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and the computer program is executed by the processor to execute the steps of the above method. .

Embodiments herein also provide computer-readable instructions, wherein when a processor executes the instructions, the program therein causes the processor to perform the methods shown in FIGS. 2 to 10 .

It should be understood that, in the various embodiments herein, the size of the sequence numbers of the above-mentioned processes does not mean the sequence of execution, and the execution sequence of each process should be determined by its functions and internal logic, rather than the implementation of the embodiments herein. The process constitutes any qualification.

It should also be understood that, in the embodiments herein, the term "and/or" is only an association relationship for describing associated objects, indicating that there may be three kinds of relationships. For example, A and/or B can mean that A exists alone, A and B exist at the same time, and B exists alone. In addition, the character "/" in this document generally indicates that the related objects are an "or" relationship.

Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. Interchangeability, the above description has generally described the components and steps of each example in terms of function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods for implementing the described functionality for each particular application, but such implementations should not be considered beyond the scope of this document.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above may refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

In the several embodiments provided herein, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may also be electrical, mechanical or other forms of connection.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solutions in the embodiments herein.

In addition, each functional unit in each of the embodiments herein may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.

The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions in this article are essentially or part of contributions to the prior art, or all or part of the technical solutions can be embodied in the form of software products, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments herein. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

The principles and implementations of this paper are described by using specific examples in this paper. The descriptions of the above examples are only used to help understand the methods and core ideas of this paper; , there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this article.

Claims (10)

1. A method for evaluating the quality of a railway track, comprising:

acquiring the detection data of the track geometry of the railway track;

importing the detection data into a predictive model, and determining a vehicle response of the railway track;

determining a track quality of the railway track based on the vehicle response.

2. The method of claim 1, wherein prior to importing the test data into a predictive model to determine a vehicle response of the railroad track, comprising:

performing wavelet decomposition on the detection data to obtain a plurality of wavelength components;

classifying a plurality of wavelength components according to wavelength ranges;

introducing several types of said wavelength components into said predictive model;

wherein, the track geometry comprises a left high and low, a right high and low, a left track direction, a right track direction, a track gauge, a horizontal pit and a triangular pit.

3. The method of claim 1, wherein the predictive model is obtained by a training method comprising:

selecting a training subset from a training set and inputting the training subset into the prediction model to obtain a vehicle response prediction value corresponding to the wavelength component;

determining the mean square error loss of the training subset according to the vehicle response predicted value and the vehicle response real value corresponding to the training subset;

obtaining regularization loss according to the regularization coefficient, the model parameters and the mean square error loss;

sequentially calculating partial derivatives of the regularization loss with respect to hidden states of the prediction model in an inverse time series;

updating the model parameters according to the partial derivatives, the hidden states, the model parameters and a preset learning rate of the prediction model;

and repeating the steps until the preset training step number or the reduction rate of the regularization loss is smaller than a preset transformation threshold value, and stopping the training of the prediction model.

4. The method of claim 3, further comprising, after stopping the training:

verifying whether the prediction model meets a verification condition, and determining that the training of the prediction model is finished when the prediction model meets the verification condition;

the verification conditions include:

determining whether the prediction result of the prediction model is accurate or not according to a comparison image drawn by the training set and the verification set in the regularization loss of each training step and a comparison image drawn by the prediction model stopped from training in the verification set in the predicted vehicle response and the real vehicle response;

if not, expanding the training set and retraining the prediction model;

and if so, determining that the prediction model is successfully trained.

5. The method of claim 4, wherein the determining whether the prediction results of the predictive model are accurate further comprises:

evaluating whether the prediction result of the prediction model stopping training is accurate or not according to a root mean square error formula and a Hill inequality coefficient formula;

the root mean square error formula is

The coefficient of Hill inequality is as follows

Wherein, yi and

respectively representing real value y and predicted value of vehicle response in verification set

I ═ 1,2, …, M is the validation set capacity;

and if the result obtained by the root mean square error formula meets a first error threshold value and the result obtained by the Hill-Eiff coefficient formula meets a second error threshold value, judging that the prediction result of the prediction model is accurate.

6. The method according to claim 3, wherein before selecting a training subset from the training set and inputting the training subset to the predictive model to obtain the predicted value of the vehicle response corresponding to the wavelength component, the method comprises:

according to a normalized formula

Determining the track geometry and normative data for each dimension of the vehicle response, resulting in normative data, where x represents the track geometry for a dimension and σ represents the standard deviation of the track geometry for that dimension,

a mean value representing the trajectory geometry in that dimension;

selecting a plurality of sections from the normalized data as the training set and the verification set;

initializing the model parameters and the hidden states with random numbers to obtain the predictive model.

7. The method of assessing railway track quality as claimed in claim 1, wherein said determining said railway track quality for each section based on said vehicle response further comprises:

calculating a comfort index and a safety index of the railway track of each section according to the vehicle response;

evaluating the comfort index and the safety index according to a limit value to determine the quality of the railway track of each section.

8. The method of claim 7, wherein the calculating a comfort index and a safety index for the railway track for each section based on the vehicle response further comprises:

the vehicle response comprises a vehicle body vertical acceleration, a vehicle body transverse acceleration, a left wheel vertical force, a right wheel vertical force, a left wheel transverse force and a right wheel transverse force;

the comfort level index comprises a vertical acceleration and a transverse acceleration of the vehicle body;

the safety indexes comprise derailment coefficients, wheel load shedding rates and transverse forces of wheel shafts;

calculating a derailment coefficient according to the left wheel transverse force, the right wheel transverse force, the left wheel vertical force and the right wheel vertical force;

calculating a wheel weight load shedding rate according to the load shedding amount of the wheel rail vertical force relative to the average static wheel weight and the average static wheel weight, wherein the average static wheel weight is the average value of the static lower left wheel vertical force and the static lower right wheel vertical force;

calculating the axle lateral force from the left wheel lateral force and the right wheel lateral force.

9. The method of claim 8, wherein the evaluating the comfort index and the safety index against limits to determine the railway track quality for each section further comprises:

the limits include a comfort limit and a security limit;

the comfort limit comprises a first limit and a second limit;

the safety limits include a third limit, a fourth limit, and a fifth limit;

determining whether the vertical acceleration of the vehicle body exceeds the first limit value, determining whether the lateral acceleration of the vehicle body exceeds the second limit value, and if not, determining that the comfort index of the section reaches the standard;

determining whether the derailment coefficient exceeds a third limit value, determining whether the wheel load shedding rate exceeds a fourth limit value, determining whether the wheel axle transverse force exceeds a fifth limit value, and if not, determining that the safety index of the section reaches the standard;

taking the section with the comfort index and the safety index reaching the standard as a section with good railway track quality;

and taking the section with at least one substandard comfort index and safety index as the section with poor railway track quality.

10. A railroad track quality evaluation device, comprising:

the acquisition unit is used for acquiring the detection data of the track geometry of the railway track;

the leading-in unit is used for leading the detection data into a prediction model and determining the vehicle response of the railway track;

a determination unit for determining a track quality of the railway track from the vehicle response.

Images