Turnout operation state rapid detection method based on traction force envelope line

Abstract

The invention belongs to the technical field of intelligent operation and maintenance of rail transit, and discloses a turnout operation state rapid detection method based on a traction force envelope line. The method adopted by the invention comprises the following steps: if a traction curve to be detected is in an upper envelope line and a lower envelope line of a normal reference curve, the detection result is normal; and if the traction curve to be detected exceeds the envelope line, determining that the detection result is matched curve information obtained by utilizing the DTW algorithm, and further updating the upper and lower envelope lines of the normal reference curve and writing the upper and lower envelope lines into a curve sample library when the matching result is normal. According to the method adopted by the invention, whether the traction curve is normal or not can be quickly determined, so that a large amount of invalid DTW distance calculation is avoided, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of intelligent operation and maintenance of rail transit, and in particular to a method for quickly detecting the operating status of a turnout based on a traction force envelope. Background Art

[0002] Turnouts are one of the basic equipment in the railway signal system, which directly affects the driving efficiency and safety. Using the collected data to detect the operating status of turnouts has important practical value. At present, the data collection sources of turnout switch machines mainly include gap images, operating current, power and traction. Compared with other monitoring data, the traction curve can directly reflect the resistance and mechanical properties of the turnout during the operation process.

[0003] Dynamic Time Warping (DTW) is an algorithm based on dynamic programming strategy to adjust two time series nonlinearly in the time domain in order to correctly calculate the similarity between the two. It does not require the length of each sequence to be strictly aligned, and allows stretching or compression on the time axis. The traditional steps of detecting the turnout state based on the DTW algorithm are: calculating the DTW distance between the curve to be detected and each sample curve in the sample curve library, and the detection result is the state information carried by the sample curve corresponding to the minimum DTW distance. However, the DTW algorithm has high time complexity, and the actual turnout is in normal operation most of the time, resulting in the problem of calculating a large number of invalid results and a long time. Summary of the invention

[0004] The purpose of the present invention is to provide a method for quickly diagnosing the operation status of a turnout based on a traction force envelope, and to address the deficiencies of existing turnout fault diagnosis methods, so as to quickly determine whether the traction force curve is normal, thereby avoiding a large number of invalid DTW distance calculations and improving detection efficiency.

[0005] To achieve the above object, the present invention provides a method for quickly diagnosing the operation status of a turnout based on a traction force envelope, comprising the following steps: Step 1: extract the upper and lower envelopes of the normal reference curve from the traction force curve sample library to obtain the threshold of the normal reference curve; Step 2: Determine whether the number of data points of the traction curve to be detected exceeds the threshold of the normal reference curve. If so, proceed to step 4; otherwise, proceed to step 3. Step 3: Determine whether the traction force curve to be detected is within the upper and lower envelopes of the normal reference curve. If so, the detection result is that the traction force curve to be detected is normal and the process ends. Otherwise, proceed to step 4. Step 4: Use the optimized DTW algorithm to match the traction force curve to be detected with the sample curve in the traction force curve sample library; Step 5: Determine whether the matching result between the traction force curve to be detected and the sample curve in the traction force curve sample library is that the traction force curve to be detected is normal. If so, the detection result is that the traction force curve to be detected is normal, and then update the upper and lower envelopes of the normal reference curve in the traction force curve sample library and end the process; Otherwise, the detection result matches the state information carried by the traction force curve to be detected and the process ends.

[0006] Furthermore, in step 1, the ability of the upper and lower envelopes to screen the normal reference curve is positively correlated with the threshold value; The normal reference curve is the standard traction curve with a length of n under normal operation of the turnout. F 0, the initialization construction of its upper and lower envelopes combines the amplitude and time shift deviation, and is dynamically updated during the turnout operation. The initialization construction method of its upper and lower envelopes is: ,in U 0. L 0 is the initialization structure of the upper and lower envelopes, N U , N C The constants that control the upper and lower envelope amplitudes are used to increase the margin on the envelope amplitude. r It is used to control the constant of the time offset window to increase the margin of the envelope time shift deviation.

[0007] Furthermore, in step 2, the threshold is set based on the fluctuation range of the number of data points of the normal reference curve, while increasing the margin; The ability to compare the number of data points to prune the curve under abnormal operating conditions is negatively correlated with the threshold value. If the number of data points of the traction force curve to be detected obviously exceeds the fluctuation range of the number of data points of the normal reference curve, the traction force curve to be detected is in an abnormal operating state and does not need to be compared with the envelope.

[0008] Furthermore, in step 3, if the length of the traction force curve to be detected is less than or equal to the length of the envelope, then each data point of the traction force curve to be detected is compared one by one to see whether it is between the corresponding points of the upper and lower envelopes; If the length of the traction force curve to be detected is greater than the length of the envelope, the data points exceeded by the traction force curve to be detected are compared with the last data points of the upper and lower envelopes.

[0009] Furthermore, in step 3, for the upper envelope U and the lower envelope L with a length of n and the traction force curve F to be detected with a length of m, if the traction force curve F to be detected is between the upper envelope U and the lower envelope L, then the formula is satisfied: .

[0010] Furthermore, in step 4, optimizing the DTW algorithm includes an early termination method, a lower bound function method, and a method for adjusting the calculation order of lower bound function data points; The process of using the optimized DTW algorithm to match the traction force curve F to be detected with the sample curve in the traction force curve sample library includes: S1: The descending sorting index value of the absolute value calculated after the Z-score normalization of the traction force curve to be tested; S2: Initialize the matching curve to the first sample curve in the curve library, and initialize the DTW minimum distance to infinity; S3: If all sample curves in the curve sample library are traversed, the process ends; otherwise, each sample curve is taken out in sequence; S4: Calculate the lower bound function of the sample curve. If the length of the traction force curve to be detected is the same as that of the sample curve, the lower bound function type is LB_Keogh, otherwise the type is LB_Yi; S5: Calculate the DTW lower bound distance between the curve to be detected and the lower bound function in descending index order. If the distance is greater than the DTW minimum distance during the calculation process, jump to step S3, otherwise go to step S6; S6: Calculate the DTW distance between the curve to be detected and the sample curve. If the distance is greater than the DTW minimum value, jump to step S3. Otherwise, the minimum value is updated to the DTW distance, the matching curve is updated to the sample curve, and jump to step S3.

[0011] Further, in step S1, for the mean m , the variance is s sequence x 1, x 2,…, x n , the sequence after Z-score standardization y 1, y 2,…, y n The calculation formula is: .

[0012] Further, in step S2, the sample curves in the sample curve library include historical fault curves, and the historical fault curves include six categories: unlocking difficulty fault curves caused by turnout close contact, point rail rebound fault curves caused by point rail deformation, conversion jam fault curves caused by foreign body intrusion, conversion difficulty fault curves caused by overtightening of point rail bolts, conversion force imbalance fault curves caused by spring aging, and locking jam fault curves caused by loose action nuts. It is determined whether there is a new fault type in the historical fault curves that does not belong to the curve sample library. If so, the sample curve library is updated.

[0013] Furthermore, in step S4, LB_Keogh and LB_Yi are two types of lower bound functions currently available: If the length of the traction force curve to be detected is the same as the length of the envelope, the LB_Keogh function constructs one of the sequences close to the upper and lower envelopes according to the sliding time window to realize the scaling calculation of the DTW distance; If the length of the traction force curve to be detected is different from the length of the envelope, the LB_Yi function selects the maximum and minimum values ​​of one of the sequences as a loose envelope to implement scaling calculation.

[0014] Further, in step S5, the updating of the upper and lower envelopes of the normal reference curve is combined with the amplitude and the time shift deviation, including the following steps: S7: For the upper envelope exceeding the original length n U , lower envelope L The normal traction force F is first calculated according to the formula Update amplitude information; where n for U , L Length, m for F Length; S8: Based on the amplitude update, according to the formula Perform time-shift deviation update; record the original U , L The index set of updated data points in index U , index L ,in r To control the constant of the time offset window, the size is set the same as in step 1.

[0015] Beneficial effects: The present invention provides a method for quickly detecting the operation status of a turnout based on a traction force envelope. Specifically, if the traction force curve to be detected is within the upper and lower envelopes of a normal reference curve, the detection result is normal; if the traction force curve to be detected exceeds the envelope, the detection result is the matching curve information obtained by using the DTW algorithm. At the same time, for the case where the matching result is normal, the upper and lower envelopes of the normal reference curve are further updated and written into the curve sample library. According to the method disclosed in the present invention, it is possible to quickly determine whether the traction force curve is normal, thereby avoiding a large number of invalid DTW distance calculations and improving detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a general flow chart of turnout state detection involved in an embodiment of the present invention; Figure 2 is a flow chart of an optimized DTW algorithm involved in an embodiment of the present invention; Figure 3 Schematic diagram of the LB_Yi lower bound function involved in an embodiment of the present invention; Figure 4 is a schematic diagram of the LB_Keogh lower bound function involved in an embodiment of the present invention; Figure 5 is a normal reference curve and its initialization upper and lower envelope diagrams involved in the embodiment of the present invention; Figure 6 is a schematic diagram of a detection result according to an embodiment of the present invention, in which the traction force curve to be detected is normal but exceeds an initial envelope; Figure 7 is a schematic diagram of updating the initial envelope amplitude involved in an embodiment of the present invention; Figure 8 is a schematic diagram of updating the initial envelope time offset involved in an embodiment of the present invention; Fig. 9 It is a curve sample library diagram involved in the embodiment of the present invention. DETAILED DESCRIPTION

[0018] The preferred mechanism and implementation method of the present invention are further described below in conjunction with the accompanying drawings and specific implementation methods.

[0019] like Figures 1 to 9 As shown, an embodiment of the present invention discloses a technical solution for a method for quickly detecting the operating status of a turnout based on a traction force envelope. Example 1

[0020] The present invention provides a method for quickly detecting the operation status of a turnout based on a traction force envelope. The overall flow chart of the turnout status detection is as follows: Figure 1 As shown, the following steps are included: Step 1: extract the upper and lower envelopes of the normal reference curve from the traction force curve sample library to obtain the threshold of the normal reference curve; Step 2: Determine whether the number of data points of the traction curve to be detected exceeds the threshold of the normal reference curve. If so, proceed to step 4; otherwise, proceed to step 3. Step 3: Determine whether the traction force curve to be detected is within the upper and lower envelopes of the normal reference curve. If so, the detection result is that the traction force curve to be detected is normal and the process ends. Otherwise, proceed to step 4. Step 4: Use the optimized DTW algorithm to match the traction force curve to be detected with the sample curve in the traction force curve sample library; Step 5: Determine whether the matching result between the traction force curve to be detected and the sample curve in the traction force curve sample library is that the traction force curve to be detected is normal. If so, the detection result is that the traction force curve to be detected is normal, and then update the upper and lower envelopes of the normal reference curve in the traction force curve sample library and end the process; Otherwise, the detection result matches the state information carried by the traction force curve to be detected and the process ends. Example 2

[0021] The steps of this embodiment are basically the same as those of embodiment 1, except that in step 1, the normal reference curve is a standard traction force curve with a length of n under normal operation of the turnout. F 0, the initialization construction of its upper and lower envelopes combines the amplitude and time shift deviation, and is dynamically updated during the turnout operation. The initialization construction method of its upper and lower envelopes is: ,in U 0. L 0 is the initialization structure of the upper and lower envelopes, N U , N C The constants that control the upper and lower envelope amplitudes are used to increase the margin on the envelope amplitude. r To control the constant of the time offset window, it is used to increase the margin of the time shift deviation of the envelope. It should be noted that the ability of the envelope to filter normal curves is positively correlated with the threshold, but a too large threshold may cause abnormal curves to be misdetected as normal classifications. Since the envelope is dynamically updated during operation, a more robust smaller value can be selected during initialization.

[0022] In this embodiment, the normal reference curve F The sampling interval of 0 is 10ms, and the number of points is n , then the construction method of initializing the upper and lower envelopes is as follows: Here, the amplitude of the upper and lower envelopes are set to a traction force of 900N, and the time window is set to 20 points before and after, that is, 0.2s before and after.

[0023] In step 3, for the upper envelope U and the lower envelope L with a length of n and the traction force curve F to be detected with a length of m, it is indicated when the following formula is satisfied: F Between the envelopes: ; In step 4, the optimization of the DTW algorithm includes an early termination method, a lower bound function method, and a method for adjusting the calculation order of the lower bound function data points; The process of using the optimized DTW algorithm to match the traction force curve F to be detected with the sample curves in the traction force curve sample library is as follows: Figure 2 As shown, the following steps are included: S1: The descending sorting index value of the absolute value calculated after the Z-score normalization of the traction force curve to be tested; S2: Initialize the matching curve to the first sample curve in the curve library, and initialize the DTW minimum distance to infinity; S3: If all sample curves in the curve sample library are traversed, the process ends; otherwise, each sample curve is taken out in sequence; S4: Calculate the lower bound function of the sample curve. If the length of the traction force curve to be detected is the same as that of the sample curve, the lower bound function type is LB_Keogh, otherwise the type is LB_Yi; S5: Calculate the DTW lower bound distance between the curve to be detected and the lower bound function in descending index order. If the distance is greater than the DTW minimum distance during the calculation process, jump to step S3, otherwise go to step S6; S6: Calculate the DTW distance between the curve to be detected and the sample curve. If the distance is greater than the DTW minimum value, jump to step S3. Otherwise, the minimum value is updated to the DTW distance, the matching curve is updated to the sample curve, and jump to step S3.

[0024] Specifically, in step S1, for the sequence x 1, x 2,…, x n , whose mean is m , the variance is s , then the sequence after Z-score standardization y 1, y 2,…, y n The calculation formula is: ; Specifically, in step S2, the sample curves in the sample curve library include historical fault curves, and the historical fault curves include six categories: unlocking difficulty fault curves caused by turnout close fitting, point rail rebound fault curves caused by point rail deformation, conversion jam fault curves caused by foreign body intrusion, conversion difficulty fault curves caused by overtightening of point rail bolts, conversion force imbalance fault curves caused by spring aging, and locking jam fault curves caused by loose action nuts. Due to the variety of turnout fault types, the sample curves in the curve sample library are difficult to include all fault conditions, and need to be updated during operation, for example, through regular inspections by professional operation and maintenance personnel or cluster analysis, to determine whether there are new fault types in the historical fault curves that do not belong to the curve sample library. If so, the sample curve library is updated.

[0025] Specifically, in step S4, LB_Keogh and LB_Yi are two types of lower bound functions currently available, such as Figure 3 and Figure 4As shown in the figure, the LB_Keogh function constructs the upper and lower envelopes of one of the sequences according to the sliding time window to realize the scaling calculation of the DTW distance. The calculation result is more accurate but requires the two sequences to be of the same length. The LB_Yi function selects the maximum and minimum values ​​of one of the sequences as the loose envelope to realize the scaling calculation. The calculation result is less accurate but has no requirement on the sequence length.

[0026] Specifically, in step S5, the traditional lower limit distance calculation order is from left to right, but the data of the actual turnout in the initial unlocking stage is small, and the impact on the lower limit distance is low, while the data points with a higher impact on the lower limit distance are concentrated in the middle conversion stage. Therefore, the descending index obtained in step S1 is used to correct the calculation process to reduce the amount of calculation.

[0027] Further, in step 5, for the U , lower envelope L Normal traction F , first update the amplitude information according to the following formula: ; in n for U , L Length, m for F Length. U , L The index set of updated data points in index U , index L , then the formula for updating the time-shift deviation based on the amplitude update is: ; in r To control the constant of the time offset window, the size is set the same as in step 1.

[0028] Figures 5 to 8 is the initialization envelope updating process of the normal reference curve in the specific implementation mode, Figure 5 is the normal reference curve and its initialization upper and lower envelope graphs; Figure 6 is a schematic diagram showing that the detection result is that the traction force curve to be detected is normal but exceeds the initial envelope; Figure 7 It is a schematic diagram of the initial envelope amplitude update; Figure 8 It is a schematic diagram of the initial envelope time offset update involved in an embodiment of the present invention. Example 3

[0029] The steps of this embodiment are basically the same as those of Embodiment 2, except that the pulling-in traction data of a subway line, a station, and a switch machine in December 2023 and January 2024 are used as an example to further explain and illustrate the method proposed in this patent.

[0030] The turnout switch machine has 3 sets of abnormal data and 1070 sets of normal data in December 2023, and 3 sets of abnormal data and 2371 sets of normal data in January 2024. The data in December 2023 are used to construct the sample curve library and the upper and lower envelopes of the normal reference curve, such as Fig. 9 shown.

[0031] A .py program was written in Python to compare the efficiency of the method proposed in this patent and the traditional DTW method. The operating hardware environment is as follows: CPU is R7-5800H, GPU is RTX3060, memory is 16GB, and Python version is 3.10. The traditional method of traversing and calculating the DTW distance between all sample curves takes 130.09 seconds, while the envelope-based method proposed in this patent takes 3.41 seconds. Among the 2371 sets of normal data, 2327 sets of data are within the envelope, and only 44 sets of data need to calculate the DTW distance, which greatly improves the detection efficiency. In addition, the three sets of abnormal data used for testing are all matched with the fault sample curve 3. The information carried by the matching curve is difficult to convert, which is consistent with the actual situation. At the same time, the normal data is not misclassified as an abnormal curve, that is, the detection accuracy is 100%. The present invention provides a method for quickly detecting the operation state of a turnout based on a traction force envelope. Specifically, if the traction force curve to be detected is within the upper and lower envelopes of a normal reference curve, the detection result is normal; if the traction force curve to be detected exceeds the envelope, the detection result is the matching curve information obtained by using a DTW algorithm. Meanwhile, for the case where the matching result is normal, the upper and lower envelopes of the normal reference curve are further updated and written into a curve sample library. According to the method disclosed in the present invention, it is possible to quickly determine whether the traction force curve is normal, thereby avoiding a large number of invalid DTW distance calculations and improving the detection efficiency.

[0032] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. However, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for quickly detecting the running state of a turnout based on the traction force envelope, characterized in that: The following steps are involved: Step 1: extract the upper and lower envelopes of the normal reference curve from the traction force curve sample library to obtain the threshold of the normal reference curve; Step 2: Determine whether the number of data points of the traction curve to be detected exceeds the threshold of the normal reference curve. If so, proceed to step 4; otherwise, proceed to step 3. Step 3: Determine whether the traction force curve to be detected is within the upper and lower envelopes of the normal reference curve. If so, the detection result is that the traction force curve to be detected is normal and the process ends. Otherwise, proceed to step 4. Step 4: Use the optimized DTW algorithm to match the traction force curve to be detected with the sample curve in the traction force curve sample library; Step 5: Determine whether the matching result between the traction force curve to be detected and the sample curve in the traction force curve sample library is that the traction force curve to be detected is normal. If so, the detection result is that the traction force curve to be detected is normal, and then update the upper and lower envelopes of the normal reference curve in the traction force curve sample library and end the process; Otherwise, the detection result matches the state information carried by the traction force curve to be detected and the process ends.

2. The method for rapid detection of turnout operation status based on traction envelope according to claim 1 is characterized in that: In step 1, the ability of the upper and lower envelopes to screen the normal reference curve is positively correlated with the threshold value; The normal reference curve is the standard traction curve with a length of n under normal operation of the turnout. F 0, the initialization construction of its upper and lower envelopes combines the amplitude and time shift deviation, and is dynamically updated during the turnout operation. The initialization construction method of its upper and lower envelopes is: .

3. The method for rapid detection of turnout operation status based on traction envelope according to claim 1 is characterized in that: In step 2, the threshold is set based on the fluctuation range of the number of data points of the normal reference curve, while increasing the margin; The ability to compare the number of data points to prune the curve under abnormal operating conditions is negatively correlated with the threshold value. If the number of data points of the traction force curve to be detected obviously exceeds the fluctuation range of the number of data points of the normal reference curve, the traction force curve to be detected is in an abnormal operating state and does not need to be compared with the envelope.

4. The method for rapid detection of turnout operation status based on traction envelope according to claim 1 is characterized in that: In step 3, if the length of the traction force curve to be detected is less than or equal to the length of the envelope, then each data point of the traction force curve to be detected is compared one by one to see whether it is between the corresponding points of the upper and lower envelopes; If the length of the traction force curve to be detected is greater than the length of the envelope, the data points exceeded by the traction force curve to be detected are compared with the last data points of the upper and lower envelopes.

5. The method for rapid detection of turnout operation status based on traction envelope according to claim 4 is characterized in that: In step 3, for the upper envelope U and the lower envelope L with a length of n and the traction force curve F to be detected with a length of m, if the traction force curve F to be detected is between the upper envelope U and the lower envelope L, then the formula is satisfied: 。 6. The method for rapid detection of turnout operation status based on traction envelope according to claim 1 is characterized in that: In step 4, the optimization of the DTW algorithm includes an early termination method, a lower bound function method, and a method for adjusting the calculation order of the lower bound function data points; The process of matching the to-be-detected traction force curve F with the sample curves in the traction force curve sample library using the optimized DTW algorithm includes: S1: The descending sorting index value of the absolute value calculated after the Z-score normalization of the traction force curve to be tested; S2: Initialize the matching curve to the first sample curve in the curve library, and initialize the DTW minimum distance to infinity; S3: If all sample curves in the curve sample library are traversed, the process ends; otherwise, each sample curve is taken out in sequence; S4: Calculate the lower bound function of the sample curve. If the length of the traction force curve to be detected is the same as that of the sample curve, the lower bound function type is LB_Keogh, otherwise the type is LB_Yi; S5: Calculate the DTW lower bound distance between the curve to be detected and the lower bound function in descending index order. If the distance is greater than the DTW minimum distance during the calculation process, jump to step S3, otherwise go to step S6; S6: Calculate the DTW distance between the curve to be detected and the sample curve. If the distance is greater than the DTW minimum value, jump to step S3. Otherwise, the minimum value is updated to the DTW distance, the matching curve is updated to the sample curve, and jump to step S3.

7. The method for quickly detecting the operation status of a turnout based on the traction force envelope according to claim 6, characterized in that: In step S1, for the mean m , the variance is s sequence x 1, x 2,…, x n , the sequence after Z-score standardization y 1, y 2,…, y n The calculation formula is: .

8. The method for quickly detecting the operation status of a turnout based on the traction force envelope according to claim 6, characterized in that: In step S2, the sample curves in the sample curve library include historical fault curves, which include six categories: unlocking difficulty fault curves caused by turnout close contact, point rail rebound fault curves caused by point rail deformation, conversion jam fault curves caused by foreign body intrusion, conversion difficulty fault curves caused by overtightening of point rail bolts, conversion force imbalance fault curves caused by spring aging, and locking jam fault curves caused by loose action nuts. It is determined whether there is a new fault type in the historical fault curves that does not belong to the curve sample library. If so, the sample curve library is updated.

9. The method for quickly detecting the operation status of a turnout based on the traction force envelope according to claim 6, characterized in that: In step S4, LB_Keogh and LB_Yi are two types of lower bound functions currently available: If the length of the traction force curve to be detected is the same as the length of the envelope, the LB_Keogh function constructs one of the sequences close to the upper and lower envelopes according to the sliding time window to realize the scaling calculation of the DTW distance; If the length of the traction force curve to be detected is different from the length of the envelope, the LB_Yi function selects the maximum and minimum values ​​of one of the sequences as a loose envelope to implement scaling calculation.

10. The method for quickly detecting the operation status of a turnout based on the traction force envelope according to claim 6, characterized in that: In step S5, the updating of the upper and lower envelopes of the normal reference curve is combined with the amplitude and the time shift deviation, including the following steps: S7: For the upper envelope exceeding the original length n U , lower envelope L The normal traction force F is first calculated according to the formula Update amplitude information; S8: Based on the amplitude update, according to the formula Perform time-shift bias update.

Images