Methods, systems and storage media for diagnosing and assessing locomotive coupler instability

Abstract

This invention provides a method, system, and storage medium for diagnosing and evaluating locomotive coupler instability, belonging to the field of coupler instability diagnosis technology. The method includes: filtering and denoising historical longitudinal impulse monitoring data from existing heavy-haul locomotives to obtain filtered and denoised longitudinal impulse monitoring data; obtaining data from a limited number of measuring points; constructing a multi-channel identification model for the longitudinal impulse of the heavy-haul locomotive and obtaining multi-channel identification results; constructing a coupler instability diagnosis and evaluation model based on the multi-channel identification results; and diagnosing the coupler instability state of the target locomotive based on the coupler instability diagnosis and evaluation model and outputting instability evaluation results. This invention enables multi-channel indirect identification of the longitudinal impulse of heavy-haul locomotives under limited measuring point conditions and simultaneously completes the diagnosis and evaluation of the locomotive's operating status.

Description

Technical Field

[0001] This invention relates to the field of coupler instability diagnosis technology, specifically to a locomotive coupler instability diagnosis and assessment method, a locomotive coupler instability diagnosis and assessment system, and a storage medium. Background Technology

[0002] In recent years, with the rapid development of heavy-haul trains towards higher axle loads and longer formations, their freight transport efficiency has significantly improved. However, at the same time, the complex service environment of trains and the longitudinal impulse problem caused by asynchronous braking have become increasingly prominent, seriously threatening the operational safety of trains. The central slave locomotive, as a key load-bearing section of heavy-haul trains, is prone to generating huge longitudinal impact forces and inducing lateral instability risks. Therefore, existing research typically deploys longitudinal impulse monitoring systems in slave locomotives to assist train drivers and mitigate longitudinal impulse levels. Currently, the monitoring of longitudinal impulse levels in heavy-haul locomotives mainly relies on traditional direct measurement methods, namely monitoring through coupler force and coupler swing angle measuring devices. However, this has significant limitations in practical engineering applications: coupler strain gauges and coupler swing angle displacement sensors are prone to structural loosening and detachment in long-term harsh dynamic environments, leading to decreased measurement accuracy or even failure, seriously affecting the continuity and reliability of monitoring data, and potentially posing a risk to train operation safety due to sensor detachment. Therefore, it is necessary to conduct indirect measurement and intelligent diagnostic analysis of the longitudinal impulse level of heavy-haul locomotives to effectively ensure the long-term safe operation of trains.

[0003] Based on the above problems, researchers have proposed several feasible indirect monitoring schemes. These include quantitatively identifying coupler force through coupler buffer displacement and quantitatively identifying coupler sway angle using the relative displacement between the car body and the frame. However, these indirect monitoring methods still rely on multi-channel sensor data and only monitor independently in a single direction, which increases equipment deployment and maintenance costs and introduces additional safety hazards. More importantly, train dynamic response is essentially a unified spatiotemporal correlation of longitudinal and lateral coupling. Existing unidirectional monitoring modes cannot characterize this intrinsic correlation, deviating from the true laws of physical motion and severely limiting the accuracy and reliability of locomotive longitudinal impulse level diagnosis. Secondly, these methods can only monitor the mid-coupling force and mid-coupling sway angle, resulting in a limited monitoring range while still requiring a large number of sensor devices, making them inefficient. Furthermore, existing methods mostly remain at the state monitoring level and have not yet achieved locomotive instability diagnosis and assessment based on monitoring data. Therefore, developing a multi-channel intelligent diagnosis and assessment method for heavy-haul locomotive coupler instability with limited measurement points will effectively overcome the above limitations and is of great significance for improving the state identification capability and operational safety diagnosis and assessment level of heavy-haul locomotives. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and storage medium for diagnosing and assessing locomotive coupler instability, so as to at least solve the problem that it is difficult to effectively diagnose and assess the operating status of locomotive couplers under the condition of a limited number of monitoring points.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for diagnosing and evaluating locomotive coupler instability. The method includes: filtering and denoising historical longitudinal impulse monitoring data of existing heavy-duty locomotives to obtain filtered and denoised longitudinal impulse monitoring data; performing multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data; constructing a multi-channel identification model for the longitudinal impulse of heavy-duty locomotives based on the limited measurement point channel data, and using the multi-channel identification model to perform multi-channel characterization and identification on the limited measurement point channel data to obtain multi-channel identification results; wherein the multi-channel identification model for the longitudinal impulse of heavy-duty locomotives is composed of a global-local multi-scale periodic-trend decomposition module and a multi-channel spatiotemporal integration fusion module connected in series; constructing a coupler instability diagnosis and evaluation model based on the multi-channel identification results, and diagnosing the coupler instability state of the target locomotive based on the coupler instability diagnosis and evaluation model and outputting instability evaluation results.

[0006] Optionally, the historical longitudinal impulse monitoring data of existing heavy-duty locomotives is filtered and denoised to obtain filtered and denoised longitudinal impulse monitoring data. This includes: dividing the historical longitudinal impulse monitoring data into coupler force monitoring data and displacement monitoring data according to the type of monitored physical quantity; setting corresponding low-pass filter cutoff frequencies for the coupler force monitoring data and the displacement monitoring data based on their respective spectral characteristics; performing low-pass filtering on the coupler force monitoring data using the low-pass filter cutoff frequency corresponding to the coupler force monitoring data; performing low-pass filtering on the displacement monitoring data using the low-pass filter cutoff frequency corresponding to the displacement monitoring data; and performing time alignment and data integration on the filtered coupler force monitoring data and the filtered displacement monitoring data to form filtered and denoised longitudinal impulse monitoring data.

[0007] Optionally, multi-source channel correlation analysis is performed on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data. This includes: dividing the filtered and denoised longitudinal impulse monitoring data according to monitoring channels to form multiple longitudinal impulse monitoring channel data sequences; calculating the correlation coefficient between any two monitoring channels using the Pearson correlation coefficient based on the longitudinal impulse monitoring channel data sequences to obtain the correlation distribution results between each monitoring channel; calculating the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive's longitudinal impulse level based on the correlation distribution results; and selecting monitoring channels whose correlation meets preset conditions from the multiple longitudinal impulse monitoring channels according to the correlation strength to provide limited measurement point channel data.

[0008] Optionally, based on the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive longitudinal impulse level is calculated, including: selecting a correlation coefficient sequence corresponding to the target monitoring channel from the correlation distribution results, wherein the correlation coefficient sequence consists of the correlation coefficients between the target monitoring channel and the other monitoring channels; and determining the correlation strength of each monitoring channel relative to the target monitoring channel based on the correlation coefficient sequence, so as to characterize the response degree of the corresponding monitoring channel to changes in the locomotive longitudinal impulse level.

[0009] Optionally, constructing a multi-channel identification model for the longitudinal impulse of a heavy-duty locomotive based on the limited measurement point channel data includes: acquiring the limited measurement point channel data and using it as input data for the multi-channel identification model; constructing a global-local multi-scale periodic-trend decomposition module based on the limited measurement point channel data to decompose the limited measurement point channel data into multi-scale periodic and trend components, obtaining corresponding periodic feature representations and trend feature representations respectively; performing spatiotemporal dependency modeling on the multi-target monitoring channel data based on the periodic-trend component modeling and combined with explicit two-step numerical integration-driven data updates, and determining the modeling result as the output result of the multi-channel spatiotemporal integration fusion module; and determining the output result of the multi-channel spatiotemporal integration fusion module as the output result of the multi-channel identification model for the longitudinal impulse of a heavy-duty locomotive, thereby completing the construction of the multi-channel identification model for the longitudinal impulse of a heavy-duty locomotive.

[0010] Optionally, based on the periodic feature representation and the trend feature representation, spatiotemporal dependency modeling is performed on the multi-target monitoring channel data through periodic-trend component modeling and combined with explicit two-step numerical integration-driven data updates. The modeling result is then determined as the output of the multi-channel spatiotemporal integration fusion module. This includes: constructing a periodic-trend component modeling hybrid sub-module of the multi-channel spatiotemporal integration fusion module based on the periodic feature representation and the trend feature representation; and performing component modeling processing to weaken nonlinear interference on the periodic feature representation and the trend feature representation through the periodic-trend component modeling hybrid sub-module to obtain the periodic-trend component modeling... The modeling results of the hybrid submodule are as follows: Based on the initial target monitoring channel identification results, a reversible normalized Fourier plot submodule of the multi-channel spatiotemporal integration fusion module is constructed by driving data updates through explicit two-step numerical integration. Non-stationarity suppression and spatiotemporal dependency modeling are performed on the multi-target monitoring channel data through the reversible normalized Fourier plot submodule to obtain the modeling results of the reversible normalized Fourier plot submodule. The modeling results of the periodic-trend component modeling hybrid submodule and the modeling results of the reversible normalized Fourier plot submodule are summed, and the summation result is determined as the output result of the multi-channel spatiotemporal integration fusion module.

[0011] Optionally, constructing a coupler instability diagnosis and evaluation model based on the multi-channel identification results includes: acquiring the multi-channel identification results and using them as input data for the coupler instability diagnosis and evaluation model; constructing a state representation structure to characterize the coupler's operational stability based on the multi-channel feature information reflecting changes in the coupler's operating state in the multi-channel identification results; constructing a diagnostic discrimination structure to distinguish different coupler operational stability states based on the state representation structure; and combining the diagnostic discrimination structure with an evaluation structure to characterize the degree of instability risk to form the coupler instability diagnosis and evaluation model.

[0012] Optionally, diagnosing the coupler instability state of the target locomotive based on the coupler instability diagnosis and evaluation model and outputting instability evaluation results includes: inputting the multi-channel identification results of the target locomotive at the corresponding time into the coupler instability diagnosis and evaluation model; performing stability discrimination calculation on the multi-channel identification results based on the coupler instability diagnosis and evaluation model to obtain the coupler stability diagnosis result of the target locomotive at the current time; based on obtaining the coupler stability diagnosis result, evaluating and calculating the time evolution trend of the coupler operating state based on the coupler instability diagnosis and evaluation model to obtain the corresponding instability evaluation result; and using the coupler stability diagnosis result and the instability evaluation result as the coupler instability diagnosis and evaluation output result of the target locomotive.

[0013] A second aspect of the present invention provides a locomotive coupler instability diagnosis and assessment system, the system comprising: a preprocessing unit for filtering and denoising historical longitudinal impulse monitoring data of existing heavy-duty locomotives to obtain filtered and denoised longitudinal impulse monitoring data; a processing unit for performing multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data; a feature recognition unit for constructing a multi-channel recognition model of longitudinal impulse of heavy-duty locomotives based on the limited measurement point channel data, and using the multi-channel recognition model to perform multi-channel characterization and recognition of the limited measurement point channel data to obtain multi-channel recognition results; wherein, the multi-channel recognition model of longitudinal impulse of heavy-duty locomotives is composed of a global-local multi-scale periodic-trend decomposition module and a multi-channel spatiotemporal integration fusion module connected in series; and an instability diagnosis unit for constructing a coupler instability diagnosis and assessment model based on the multi-channel recognition results, and diagnosing the coupler instability state of the target locomotive based on the coupler instability diagnosis and assessment model and outputting instability assessment results.

[0014] On the other hand, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described locomotive coupler instability diagnosis and assessment method.

[0015] Through the above technical solution, this invention filters and reduces noise from historical longitudinal impulse monitoring data of heavy-duty locomotives and performs multi-source channel correlation analysis. Under the condition of a limited number of monitoring points, this effectively filters key monitoring channels, thus avoiding interference from redundant or weakly correlated channels in subsequent analysis results. Based on this, a multi-channel identification model is constructed, composed of a global-local multi-scale period-trend decomposition module and a multi-channel spatiotemporal integration fusion module. This model performs multi-channel characterization and identification on limited monitoring point channel data, obtaining multi-channel identification results that reflect changes in the coupler's operating state. Furthermore, a coupler instability diagnosis and evaluation model is constructed based on the multi-channel identification results. This allows for effective diagnosis and evaluation of the target locomotive's coupler instability state without the need to deploy a large number of monitoring points, thereby improving the completeness and reliability of the coupler operating state analysis.

[0016] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0018] Figure 1This is a flowchart of the steps of a locomotive coupler instability diagnosis and assessment method provided in one embodiment of the present invention;

[0019] Figure 2 This is a schematic diagram of the sensor installation position on a heavy-duty locomotive and a schematic diagram of the multi-source channel correlation analysis results provided in one embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the global-local multi-scale period-trend decomposition module structure in a multi-channel recognition model provided by one embodiment of the present invention;

[0021] Figure 4 This is a schematic diagram of the periodic-trend sub-modeling hybrid sub-module structure of the multi-channel spatiotemporal integration fusion module in the multi-channel recognition model provided by one embodiment of the present invention;

[0022] Figure 5 This is a schematic diagram of the reversible normalized Fourier plot submodule structure of the multi-channel spatiotemporal integral fusion module in a multi-channel recognition model provided by an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of a multi-channel recognition model for longitudinal impulses of heavy-duty locomotives under limited measurement points, provided by one embodiment of the present invention.

[0024] Figure 7 This is a schematic diagram of the coupler instability diagnosis and evaluation model provided in one embodiment of the present invention;

[0025] Figure 8 This is a schematic diagram of multi-channel recognition results obtained based on measured data according to one embodiment of the present invention;

[0026] Figure 9 This is a system structure diagram of a locomotive coupler instability diagnosis and evaluation system provided in one embodiment of the present invention. Detailed Implementation

[0027] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0028] Figure 1 This is a flowchart illustrating the steps of a locomotive coupler instability diagnosis and assessment method provided in one embodiment of the present invention. Figure 1 As shown, this invention provides a method for diagnosing and assessing locomotive coupler instability, the method comprising:

[0029] Step S10: Filter and denoise the historical longitudinal impulse monitoring data of existing heavy-duty locomotives to obtain filtered and denoised longitudinal impulse monitoring data.

[0030] Specifically, the historical longitudinal impulse monitoring data is divided into coupler force monitoring data and displacement monitoring data according to the type of monitored physical quantity. Based on the spectral characteristics of the coupler force monitoring data and the displacement monitoring data, corresponding low-pass filter cutoff frequencies are set for the coupler force monitoring data and the displacement monitoring data, respectively. Low-pass filtering processing is performed on the coupler force monitoring data using the low-pass filter cutoff frequency corresponding to the coupler force monitoring data. Low-pass filtering processing is also performed on the displacement monitoring data using the low-pass filter cutoff frequency corresponding to the displacement monitoring data. The filtered coupler force monitoring data and the filtered displacement monitoring data are time-aligned and integrated to form filtered and noise-reduced longitudinal impulse monitoring data.

[0031] In this embodiment of the invention, historical longitudinal impulse monitoring data of existing heavy-haul locomotives are filtered and denoised to obtain filtered and denoised longitudinal impulse monitoring data that can be used for subsequent analysis. The historical longitudinal impulse monitoring data originates from monitoring devices deployed on the couplers and related parts during locomotive operation. During the acquisition process, the monitoring data is inevitably affected by factors such as sensor noise, structural vibration interference, and changes in the operating environment. Therefore, preprocessing of the original monitoring data is necessary before proceeding to subsequent multi-source channel correlation analysis and multi-channel identification processing.

[0032] Specifically, based on the type of physical quantity corresponding to the monitoring data, the historical longitudinal impulse monitoring data is classified into coupler force monitoring data and displacement monitoring data. Since coupler force monitoring data and displacement monitoring data differ in physical meaning, signal amplitude variation characteristics, and spectral distribution characteristics, different filtering parameter settings are used for each in subsequent filtering processing.

[0033] After classifying the monitored data types, spectral characteristic analysis was performed on the coupler force monitoring data and displacement monitoring data to identify the main effective frequency ranges and noise bands contained in their respective signals. Based on the results of the spectral characteristic analysis, corresponding low-pass filter cutoff frequencies were set for the coupler force monitoring data and displacement monitoring data. The low-pass filter cutoff frequency for the coupler force monitoring data was used to retain the effective mechanical components related to longitudinal impulses while suppressing high-frequency noise interference; the low-pass filter cutoff frequency for the displacement monitoring data was used to smooth the displacement change curve and reduce the impact of measurement noise on the judgment of displacement trends.

[0034] After determining their respective low-pass filter cutoff frequencies, the coupler force monitoring data is low-pass filtered using the cutoff frequency corresponding to that of the coupler force monitoring data; simultaneously, the displacement monitoring data is low-pass filtered using the cutoff frequency corresponding to that of the displacement monitoring data. Through these separate filtering processes, the coupler force monitoring data and displacement monitoring data retain their main physical characteristics while reducing the impact of random noise and high-frequency interference on data stability.

[0035] After filtering, the filtered coupler force monitoring data and the filtered displacement monitoring data are time-aligned to eliminate time offset issues caused by differences in sampling timing. Subsequently, the time-aligned coupler force monitoring data and displacement monitoring data are integrated to form unified, denoised longitudinal impulse monitoring data, which is then used as input data for subsequent multi-source channel correlation analysis.

[0036] In one specific implementation, existing heavy-duty locomotive coupler force monitoring data and displacement monitoring data are filtered and noise-reduced using low-pass filter cutoff frequencies of 80Hz and 20Hz, respectively. The sampling frequency for both coupler force and displacement signals is 200Hz. In this embodiment, the low-pass filter is a 4th-order Butterworth filter with an accuracy of ±1dB within its bandwidth and an attenuation of 24dB / octave outside its bandwidth, down to -60dB.

[0037] Step S20: Perform multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data.

[0038] Specifically, the filtered and denoised longitudinal impulse monitoring data is divided into multiple longitudinal impulse monitoring channel data sequences according to the monitoring channels. Based on the longitudinal impulse monitoring channel data sequences, the correlation coefficient between any two monitoring channels is calculated using the Pearson correlation coefficient to obtain the correlation distribution results between each monitoring channel. Based on the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive's longitudinal impulse level is calculated. According to the correlation strength, monitoring channels whose correlation meets preset conditions are selected from the multiple longitudinal impulse monitoring channels to provide limited measurement point channel data.

[0039] Specifically, based on the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive longitudinal impulse level is calculated, including: selecting a correlation coefficient sequence corresponding to the target monitoring channel from the correlation distribution results, wherein the correlation coefficient sequence consists of the correlation coefficients between the target monitoring channel and the other monitoring channels; and determining the correlation strength of each monitoring channel relative to the target monitoring channel based on the correlation coefficient sequence, so as to characterize the response degree of the corresponding monitoring channel to changes in the locomotive longitudinal impulse level.

[0040] In this embodiment of the invention, multi-source channel correlation analysis is performed on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data for subsequent multi-channel identification. Since longitudinal impulse monitoring of heavy-haul locomotives typically involves multiple monitoring channels, and these channels differ in physical location, installation method, and signal response characteristics, directly modeling all monitoring channel data uniformly can easily introduce redundant or weakly correlated information, thus affecting the stability of the subsequent multi-channel identification process. Therefore, before constructing the multi-channel identification model, it is necessary to analyze and filter the correlation relationships between the monitoring channels.

[0041] Specifically, the filtered and denoised longitudinal impulse monitoring data is divided according to monitoring channels, forming multiple longitudinal impulse monitoring channel data sequences. Each longitudinal impulse monitoring channel data sequence corresponds to a specific monitoring channel, reflecting the longitudinal impulse changes collected by that channel during locomotive operation. Through the above channel division process, the original longitudinal impulse monitoring data is structured into multiple independently analyzable channel data sequences, providing a basis for subsequent correlation calculations.

[0042] After channel segmentation, the correlation between any two monitoring channels is calculated using the Pearson correlation coefficient based on the longitudinal impulse monitoring channel data sequences. By calculating the Pearson correlation coefficient between different monitoring channel data sequences, the degree of linear correlation between each monitoring channel in terms of temporal variation trends can be quantified. By calculating the correlation between each pair of monitoring channels, a set of correlation coefficients covering all monitoring channel combinations is obtained, thus forming a correlation distribution result reflecting the relationship between each monitoring channel.

[0043] After obtaining the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive's longitudinal impulse level is further calculated based on these results. The target monitoring channel is a reference channel in the longitudinal impulse monitoring system used to represent the overall longitudinal impulse level change, and it can be determined based on the locomotive's structural position, monitoring stability, or historical operating experience. In the specific calculation process, a correlation coefficient sequence corresponding to the target monitoring channel is selected from the correlation distribution results. This correlation coefficient sequence consists of the correlation coefficients between the target monitoring channel and the other monitoring channels.

[0044] Based on the correlation coefficient sequence, the correlation strength of each monitoring channel relative to the target monitoring channel is determined to characterize the response of the corresponding monitoring channel to changes in the locomotive's longitudinal impulse level. Through the above correlation strength calculation process, the degree of correlation between different monitoring channels during longitudinal impulse changes can be quantitatively described, thus providing a basis for the selection of monitoring channels.

[0045] After obtaining the correlation strength of each monitoring channel, monitoring channels that meet preset conditions are selected from multiple longitudinal impulse monitoring channels based on the correlation strength, and the data corresponding to the selected monitoring channels are determined as finite measurement point channel data. Through the above multi-source channel correlation analysis and screening process, while retaining monitoring channels that can effectively reflect changes in the locomotive's longitudinal impulse level, the interference of weakly correlated or redundant monitoring channels on the subsequent multi-channel identification process is reduced, thereby providing a stable data foundation for the subsequent construction of a multi-channel identification model based on finite measurement point channel data.

[0046] In one specific implementation, the Pearson correlation coefficient is used to perform multi-source channel correlation analysis on the longitudinal impulse monitoring data of heavy-haul locomotives, and the correlation results are as follows: Figure 2 As shown. The expression for the Pearson correlation coefficient is as follows:

[0047] (1)

[0048] In the formula: For variables covariance; Let X be the standard deviation of variable X; Let Y be the standard deviation of the variable. The Pearson correlation coefficient is used. and Let X and Y represent the values ​​of the i-th sample point and the i-th sample point, respectively. Let X be the average value of the variable. Let be the average value of variable Y, and n be the sample size.

[0049] It should be noted that, Figure 2 The meanings of the letters in the Chinese alphabet are as follows:

[0050] LRDCC: Relative displacement between coupler and car body; FCF / MCF / RCF: Front / Middle / Rear coupler force; RDCF-A / B / C / D: Relative displacement between the four frames and the car body; MCYA: Middle coupler swing angle; CM: Car body misalignment.

[0051] In this embodiment, the coupler force, coupler swing angle, and car body misalignment, which fully reflect the longitudinal impulse level of heavy-haul locomotives, are selected as the multi-channel identification objects. Based on the correlation results, a weighted average is calculated for the monitoring data of the above three types of channels. In the longitudinal direction, the average correlation coefficient between LRDCC and FCF, MCF, and RCF is 0.8533; in the transverse direction, the average correlation coefficients between RDCF-A / B / C / D and MCYA and CM are 0.7650, 0.8500, 0.8600, and 0.8150, respectively. Based on the above correlation strength, the LRDCC and RDCF-C channels are finally selected as the inputs for the subsequent multi-channel identification model.

[0052] Step S30: Construct a multi-channel identification model for the longitudinal impulse of a heavy-duty locomotive based on the limited measurement point channel data, and use the multi-channel identification model to perform multi-channel characterization and identification on the limited measurement point channel data to obtain multi-channel identification results.

[0053] Specifically, constructing a multi-channel identification model for the longitudinal impulse of a heavy-haul locomotive based on the limited measurement point channel data includes: acquiring the limited measurement point channel data and using it as input data for the multi-channel identification model; constructing a global-local multi-scale periodic-trend decomposition module based on the limited measurement point channel data to decompose the limited measurement point channel data into multi-scale periodic and trend components, obtaining corresponding periodic feature representations and trend feature representations respectively; performing spatiotemporal dependency modeling on the multi-target monitoring channel data based on the periodic-trend component modeling and combined with explicit two-step numerical integration-driven data updates, and determining the modeling result as the output result of the multi-channel spatiotemporal integration fusion module; and determining the output result of the multi-channel spatiotemporal integration fusion module as the output result of the multi-channel identification model for the longitudinal impulse of a heavy-haul locomotive, thereby completing the construction of the multi-channel identification model for the longitudinal impulse of a heavy-haul locomotive.

[0054] Optionally, based on the periodic feature representation and the trend feature representation, spatiotemporal dependency modeling is performed on the multi-target monitoring channel data through periodic-trend component modeling and combined with explicit two-step numerical integration-driven data updates. The modeling result is then determined as the output of the multi-channel spatiotemporal integration fusion module. This includes: constructing a periodic-trend component modeling hybrid sub-module of the multi-channel spatiotemporal integration fusion module based on the periodic feature representation and the trend feature representation; and performing component modeling processing to weaken nonlinear interference on the periodic feature representation and the trend feature representation through the periodic-trend component modeling hybrid sub-module to obtain the periodic-trend component modeling... The modeling results of the hybrid submodule are as follows: Based on the initial target monitoring channel identification results, a reversible normalized Fourier plot submodule of the multi-channel spatiotemporal integration fusion module is constructed by driving data updates through explicit two-step numerical integration. Non-stationarity suppression and spatiotemporal dependency modeling are performed on the multi-target monitoring channel data through the reversible normalized Fourier plot submodule to obtain the modeling results of the reversible normalized Fourier plot submodule. The modeling results of the periodic-trend component modeling hybrid submodule and the modeling results of the reversible normalized Fourier plot submodule are summed, and the summation result is determined as the output result of the multi-channel spatiotemporal integration fusion module.

[0055] In this embodiment of the invention, a multi-channel identification model for longitudinal impulse of heavy-duty locomotives is constructed based on the limited measurement point channel data, and the multi-channel identification model is used to perform multi-channel characterization and identification on the limited measurement point channel data to obtain multi-channel identification results.

[0056] Specifically, the limited measurement point channel data obtained through multi-source channel correlation analysis is used as the input data for the multi-channel identification model. A global-local multi-scale periodic-trend decomposition module is constructed based on this limited measurement point channel data to perform multi-scale periodic and trend component decomposition processing on the data, obtaining corresponding periodic and trend feature representations respectively. Further, based on these periodic and trend feature representations, spatiotemporal dependency modeling is performed on the multi-target monitoring channel data through periodic-trend component modeling combined with explicit two-step numerical integration-driven data updates. This constructs a multi-channel spatiotemporal integration fusion module, and the modeling result is determined as the output of this module. This output is then used as the output of the multi-channel identification model for the longitudinal impulse of heavy-duty locomotives.

[0057] In one specific implementation, the global-local multi-scale cycle-trend decomposition module is as follows: Figure 3As shown, this module first performs multi-scale decomposition of the input time series at a global scale. It extracts the corresponding periodic components through analysis windows at different time scales and uses the remaining parts as the overall trend components, thereby separating the long-term variation patterns and periodic fluctuation characteristics of the signal. Based on this, a local analysis mechanism is introduced to further refine the decomposition of periodic and trend components within a preset local time window to characterize the non-stationary variation characteristics of the signal in local intervals. During the local decomposition process, through weight allocation and weighted summation, the periodic and trend components at different scales are adaptively combined to form a decomposition result that can simultaneously reflect global evolutionary characteristics and local dynamic changes, providing structured feature input for subsequent multi-channel spatiotemporal integral fusion.

[0058] The Global-Local Multiscale Period-Trend Decomposition (GLMP-TD) process, based on the Global-Local Multiscale Period-Trend Decomposition module, is implemented through the following sub-steps:

[0059] The first step is to execute the Global Multiscale Cycle-Trend Decomposition Algorithm (GMP-TD) to decouple the global multiple micro-cycle components and macro-trend components within the data. This is based on data input from two channels: LRDCC and RDCF-C. This decomposition process is represented as follows:

[0060] (2)

[0061] (3)

[0062] In the formula: They represent the first i ( ) Column channel input data and the first i The column channel is in the first j Periodic components extracted at each scale; Represents the first value in a set of scale values. j ( ) scale values; N Indicates the global data length; M Indicates the number of input channels; n Indicates the number of scales; This represents the multi-period decomposition algorithm in the Multi-Period Trend Decomposition Method (MSTL). This indicates that the global multi-period and trend decoupling results of all channels are systematically integrated into a set. middle.

[0063] In this embodiment, the set of scale values sThe value is [256, 512]. It should be noted that the number of scales and the values ​​of each scale will affect the decomposition time and accuracy. In addition to the preferred scheme in this embodiment, other sets of scale values ​​can also be used according to the requirements of decomposition time and accuracy.

[0064] The second step is based on globally decoupled data. Within a sliding data window, the Local Multi-Scale Trend-Period Decomposition (LMT-PD) submodule aggregates local fine- and coarse-scale macro-trend information and deeply mines micro-periodic features such as longitudinal shocks and random fluctuations. For sliding data... In this regard, the process is represented as follows:

[0065] (4)

[0066] (5)

[0067] In the formula: They represent The k ( ) Column sequence and corresponding aggregated trend items; N w Indicates the length of the window data; This represents the multi-trend decomposition algorithm in the Multi-Expert Decomposition Method (MOEDecomp), which employs... Perform a moving average and use padding to keep the sequence length constant; utilize Generate data with adaptive weight distribution; This indicates element-wise multiplication; This indicates that the decoupling results of local multiple trends and cyclical fluctuations from all channels are systematically integrated into a set. middle; For the convenience of the following description, P represents the concatenated feature dimension.

[0068] In summary, using To summarize formulas (2) to (5).

[0069] In the formula, and These represent the final periodic component and the trend component of GLMP-TD, respectively.

[0070] After obtaining the periodic feature representation and the trend feature representation, a multi-channel spatiotemporal integration fusion module is constructed based on these representations. This module performs temporal correlation modeling and inter-channel fusion processing on the periodic and trend feature representations. In practice, the module processes the periodic and trend feature representations from different monitoring channels uniformly. By modeling the temporal correlations of the features from each channel and combining this with the collaborative relationships between channels, it achieves comprehensive fusion of multi-channel features in both temporal and channel dimensions. Through this spatiotemporal integration fusion process, the longitudinal impulse information contained in different monitoring channels can be integrated and expressed within a unified model structure.

[0071] In one specific implementation, the multi-channel spatiotemporal integration fusion module is implemented through the following sub-steps:

[0072] The first step involves using a hybrid period-trend modeling (PM-M) module to reduce the nonlinear complexity of the input data and effectively capture the correlations between data sequences, such as... Figure 4 As shown. The specific process is as follows:

[0073] For the dynamical periodic component, evolution is performed by stacking multiple variable-dimensional Koopman prediction (VDKoopa) submodules, specifically:

[0074] First, a Fourier filter is used ( The input data is decomposed using both time-independent and time-independent transformations, as shown in the formula:

[0075] (6)

[0076] In the formula: Indicates the first b The input periodic components of each VDKoopa block; and These represent time-invariant and time-varying periodic components, respectively.

[0077] Then, variable-dimensional time-invariant Koopman prediction was used ( ) and variable-dimensional time-varying Koopman prediction ( The mapping between time-invariant and time-varying dynamics is achieved as shown in the following formula:

[0078] (7)

[0079] In the formula, , and They represent the first bThe mapping results of non-time-varying and time-varying components and the corresponding fitted values ​​of time-varying components in each VDKoopa block; N i It is the length of the mapped output data. T This refers to the number of output channels.

[0080] Finally, multiple VDKoopa submodules are chained together, using the time-varying residual output of the current block as the input of the next block. The final mapping result of the periodic components is the sum of the outputs of all blocks. This process can be described as follows:

[0081] (8)

[0082] (9)

[0083] In the formula: This indicates the time-varying residual output of the current block; Indicates the input to the next VDKoopa block; This represents the mapping result of the final periodic components; B This indicates the total number of blocks in VDKoopa.

[0084] In this embodiment, the number of blocks of VDKoopa B The value is 2. It should be noted that the number of VDKoopa blocks affects the modeling accuracy and the number of model parameters for the periodic components. In addition to the preferred scheme in this embodiment, other values ​​can be used depending on the requirements for the modeling accuracy and the number of model parameters for the periodic components.

[0085] For the dynamic trend component, evolution is performed through a multilayer perceptron (MLP), and this process is represented as follows:

[0086] (10)

[0087] In the formula: This represents the final mapping result of the trend component. This represents two fully connected layers; This indicates a modified linear unit activation function; This indicates the discarded layer.

[0088] The mapping periodic component and trend components The combined output achieves the PM-M module outputting 5 channels. .

[0089] The second step involves mitigating data distribution drift through the Reversible Normalized Fourier Transform (RNFG) module, while simultaneously accurately capturing the spatiotemporal correlations between multiple output channels, such as... Figure 5 As shown. Given five channels: FCF, MCF, RCF, MCYA, and CM, for data input. This process is represented as follows:

[0090] First, the normalization operation of the Reversible Instance Normalization (RevIN) module is used to remove the mean of the input data. drift and variance Change, and by leveraging the learnable scale. and scaling The parameters are used for adaptive feature transformation of the data.

[0091] Then, effectively suppressing multi-channel input data Based on the distribution drift, a hypervariable graph data structure is constructed using a spatiotemporally fully connected approach, and its nodes are embedded. This process can be represented as follows:

[0092] (11)

[0093] In the formula: Each element is considered a node, such that Represent node characteristics; It is an adjacency matrix; To construct a fully connected hypervariable graph; Indicates the node embedding function; This represents the embedded node features; N w , T and d Let these represent the input data length, number of channels, and node embedding dimension, respectively; here we let This indicates the number of nodes in the hypervariable graph, for the convenience of subsequent descriptions in this paper.

[0094] Subsequently, the embedded features A Discrete Fourier Transform (DFT) is performed, and multiple Fourier Operators (FGOs) are stacked in the Fourier space to perform recursive matrix multiplication. The recursive result is then subjected to an Inverse Discrete Fourier Transform (IFT) to convert it to the time domain for output. For a given hypervariable graph Stacking Q The recursive matrix multiplication of FGOs is represented as follows:

[0095] (12)

[0096] In the formula: , and It is the identity matrix; and Let represent the adjacency structure of the q-th diffusion and the cumulative transformation of the weight matrix from 0 to q, respectively; Q is the number of Fourier plot operators. The cumulative propagation matrix from 0 to q diffusions; Describes the q-th FGO that satisfies ; It corresponds to the q-th diffusion step and has the same properties as... Same sparse pattern; , and Let represent the q-th diffusion adjacency matrix, the q-th layer weight matrix, and the q-th layer cumulative propagation matrix, respectively. It is the q-th weight matrix; It is a deviation parameter; Represents the ReLU activation function; Represents the Discrete Fourier Transform; For Fourier filter functions; Given the node feature matrix of the hypervariable graph; Given the adjacency matrix of the hypervariable graph.

[0097] In this embodiment, the number of FGOs Q The value is 3. It should be noted that the number of FGOs will affect the modeling accuracy and the number of model parameters of spatiotemporal correlation. In addition to the preferred scheme in this embodiment, other values ​​can be used according to the requirements of the modeling accuracy and the number of model parameters of spatiotemporal correlation.

[0098] Next, through MLP... By applying nonlinear processing, the spatiotemporal dependency results of the multi-channel data are finally captured. The forward propagation of an MLP can be described as follows:

[0099] (13)

[0100] Finally, for A RevIN module inverse normalization operation was performed to restore the original sequence drift characteristics, ultimately obtaining the spatiotemporal modeling results for 5 channels. ; It is a fully connected layer; For random deactivation layers; It is a linear rectification activation function.

[0101] The third step involves using an explicit two-step numerical integration method (“Zhai method”) to collaboratively fuse the PM-M and RNFG modules, yielding the final STIF model result. This method improves the ability to characterize the physical consistency of the longitudinal impulse process in heavy-haul locomotives while maintaining computational efficiency. The mathematical expression for the explicit two-step numerical integration method is:

[0102] (14)

[0103] In the formula: , and These represent the displacement, velocity, and acceleration vectors, respectively. The time integration step size; subscript n , n -1 and n +1 represents the current integration step, the previous integration step, and the next integration step, respectively; where and These are independent parameters that control the characteristics of the integration method.

[0104] After completing the multi-channel spatiotemporal integration fusion processing, the output of the multi-channel spatiotemporal integration fusion module is determined as the output structure of the multi-channel longitudinal impulse recognition model for heavy-haul locomotives, thus completing the construction of the multi-channel longitudinal impulse recognition model for heavy-haul locomotives. At this point, the multi-channel recognition model possesses the ability to perform multi-channel representation and recognition of data from limited measurement points.

[0105] In one specific implementation, the longitudinal impulse multichannel recognition model is as follows: Figure 6 As shown, the longitudinal impulse multi-channel recognition model starts with a few channel inputs. It first performs data initialization and feature preparation on the input data, then sends the initialized data to the global-local multi-scale periodic-trend decomposition module to obtain feature representations for characterizing multi-scale periodic and trend components. Subsequently, in the model iteration phase, PM-M performs mapping calculations on the periodic and trend feature representations, and RNFG models the inter-channel correlation and state evolution under explicit two-step numerical integration constraints. The two results are merged at the fusion node and input into the MLP, outputting recognition results corresponding to multiple channels, thus achieving representational recognition from a few channels to a multi-channel state.

[0106] After the multi-channel recognition model is constructed, it is further used to perform multi-channel characterization and recognition on the limited measurement point channel data to obtain multi-channel recognition results. Specifically, the limited measurement point channel data is input into the multi-channel recognition model for the longitudinal impulse of the heavy-haul locomotive; the global-local multi-scale periodic-trend decomposition module is used to decompose the limited measurement point channel data into multi-scale periodic and trend components to obtain corresponding periodic and trend feature representations; based on the periodic and trend feature representations, multi-channel temporal correlation modeling and inter-channel fusion processing are performed through the multi-channel spatiotemporal integration fusion module; and the output of the multi-channel spatiotemporal integration fusion module is determined as the multi-channel recognition result. The multi-channel recognition result is used to uniformly represent the longitudinal impulse-related state of the locomotive and serves as input data for subsequent coupler instability diagnosis and evaluation models.

[0107] Step S40: Construct a coupler instability diagnosis and evaluation model based on the multi-channel recognition results, and diagnose the coupler instability state of the target locomotive based on the coupler instability diagnosis and evaluation model and output the instability evaluation results.

[0108] Specifically, constructing a coupler instability diagnosis and evaluation model based on the multi-channel identification results includes: acquiring the multi-channel identification results and using them as input data for the coupler instability diagnosis and evaluation model; constructing a state representation structure to characterize the coupler's operational stability based on the multi-channel feature information reflecting changes in the coupler's operating state in the multi-channel identification results; constructing a diagnostic discrimination structure to distinguish different coupler operational stability states based on the state representation structure; and combining the diagnostic discrimination structure with an evaluation structure to characterize the degree of instability risk to form the coupler instability diagnosis and evaluation model.

[0109] Furthermore, based on the coupler instability diagnosis and evaluation model, the instability state of the target locomotive is diagnosed and an instability evaluation result is output. This includes: inputting the multi-channel identification result of the target locomotive at the corresponding time into the coupler instability diagnosis and evaluation model; performing stability discrimination calculation on the multi-channel identification result based on the coupler instability diagnosis and evaluation model to obtain the coupler stability diagnosis result of the target locomotive at the current time; based on the obtained coupler stability diagnosis result, evaluating and calculating the time evolution trend of the coupler operating state based on the coupler instability diagnosis and evaluation model to obtain the corresponding instability evaluation result; and using the coupler stability diagnosis result and the instability evaluation result as the coupler instability diagnosis and evaluation output result of the target locomotive.

[0110] In this embodiment of the invention, a coupler instability diagnosis and evaluation model is constructed based on the multi-channel identification results. This model is then used to diagnose the coupler instability state of the target locomotive and output instability evaluation results. The coupler instability diagnosis and evaluation model is used to further analyze the coupler operating state based on the obtained longitudinal impulse-related multi-channel state characterization, thereby enabling the determination of the coupler stability state and the assessment of the degree of instability risk.

[0111] Specifically, during the construction phase of the coupler instability diagnosis and evaluation model, multi-channel recognition results output by the longitudinal impulse multi-channel recognition model of the heavy-duty locomotive are obtained, and these multi-channel recognition results are used as input data for the coupler instability diagnosis and evaluation model. The multi-channel recognition results consist of state features corresponding to multiple channels, which are used to reflect the changes in the operating state of the coupler under longitudinal impulse.

[0112] After obtaining the multi-channel recognition results, a state representation structure for characterizing the coupler's operational stability is constructed based on the multi-channel feature information reflecting changes in the coupler's operating state. This state representation structure is used to uniformly organize and express the multi-channel feature information, enabling state features from different channels and dimensions to be comprehensively described within the same structure, thereby forming a state representation reflecting the overall operational stability of the coupler.

[0113] Based on the constructed state representation structure, a diagnostic discrimination structure is further constructed to distinguish different stable operating states of couplers. This diagnostic discrimination structure is used to classify or discriminate the operating states of couplers, distinguishing coupler states with different stability levels or different operating state categories. Through this diagnostic discrimination structure, continuously changing state representations can be mapped to discrete operating state discrimination results, providing a clear basis for subsequent instability assessment.

[0114] After constructing the diagnostic discrimination structure, it is combined with an assessment structure used to characterize the degree of instability risk to form a coupler instability diagnosis and assessment model. The assessment structure is used to quantitatively analyze the changing trends and potential risks of the coupler's operating state. Working in conjunction with the diagnostic discrimination structure, the coupler instability diagnosis and assessment model simultaneously possesses the ability to discriminate operating states and assess instability risks.

[0115] In one specific implementation, the coupler instability diagnosis and evaluation model employs a multi-branch depth-separable residual network, such as... Figure 7 As shown. The coupler instability diagnosis and assessment process is implemented through the following sub-steps:

[0116] The first step involves employing a multi-branch parallel architecture to simultaneously learn multi-dimensional sensitive features, enabling in-depth analysis of the recognition results across five channels in the multi-output channel recognition model. The depthwise separable residual module of a single branch can be described as follows:

[0117] (15)

[0118] In the formula: This is the result of a multi-channel recognition model; This represents the nonlinear transformation output obtained by concatenating depthwise separable convolutional layers (DWConv), batch normalization (BN), and ReLU activation functions; This represents a residual join (element-wise addition); It is the output feature of a single depth residual block; m This indicates a parallel branch index.

[0119] The second step is to perform adaptive feature fusion on the multi-branch outputs. The function calculates the attention weights of each branch and performs weighted fusion of the branch features to obtain the fused features. :

[0120] (16)

[0121] In the formula: These are the weight coefficients for the corresponding branches; M This represents the number of parallel branches in the deep residual structure.

[0122] In this embodiment, the number of parallel branches M The number of branches is 3, and each branch uses a 3×1 depthwise separable convolutional kernel. Although the kernel size is the same for each branch, the weights of each kernel are dynamically and adaptively adjusted during network training and inference to extract diverse feature information. Furthermore, the number of parallel branches and the size of the kernels in each branch directly affect the speed and accuracy of diagnosis and evaluation. Besides the preferred structure used in this embodiment, the above structural parameters can be flexibly adjusted according to specific requirements such as real-time performance and accuracy in practical applications. Multi-branch depthwise separable residual networks with other parameter configurations are also applicable to this invention.

[0123] The third step, based on the feature fusion results, utilizes a single-layer fully connected (FC) network and The classifier enables intelligent diagnosis and assessment of coupler instability.

[0124] In this embodiment, based on the results of multi-feature fusion, the coupler instability diagnosis and evaluation rules mainly include two aspects: first, determining the locomotive's operational stability level at the current moment, specifically divided into three levels: "stable", "critical instability" and "instability"; second, calculating the probability that the coupler force, coupler swing angle and car body misalignment key safety indicators will exceed the set threshold at the next moment.

[0125] In this embodiment, the warning thresholds for coupler force, coupler swing angle, and car body misalignment are 1000kN, 6°, and 200mm, respectively. The locomotive operation stability judgment rule is as follows: if all indicators are below 80% of the threshold, it is a "stable" state; if any indicator exceeds 80% of the threshold but is below the threshold, it is a "critical instability" state; if any indicator exceeds the threshold, it is judged as an "instability" state, at which point the control strategy needs to be adjusted immediately to suppress longitudinal impulses.

[0126] The high-quality data is divided into training, validation, and test sets. The multi-channel recognition model and the coupler instability diagnosis and evaluation model are trained and their performance evaluated to obtain the optimal model. In this embodiment, the mean squared error (MSE) is used as the model training loss function, and the AdamW optimizer is selected for parameter updates. Model performance is measured by the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 A comprehensive evaluation will be conducted.

[0127] After the coupler instability diagnosis and evaluation model is constructed, the instability state of the target locomotive is further diagnosed based on the model, and the instability evaluation result is output. Specifically, the multi-channel identification result of the target locomotive at the corresponding time is input into the coupler instability diagnosis and evaluation model. The multi-channel identification result is the state input of the target locomotive at the corresponding time, which is used to drive the diagnosis and evaluation model to perform calculations.

[0128] After inputting the multi-channel recognition results, based on the coupler instability diagnosis and evaluation model, stability discrimination calculation is performed on the multi-channel recognition results to obtain the coupler stability diagnosis result of the target locomotive at the current moment. The coupler stability diagnosis result is used to reflect the stability category of the coupler operating state of the target locomotive at the current moment.

[0129] Based on the obtained coupler stability diagnosis results, the time evolution trend of the coupler's operating state is further evaluated and calculated using the coupler instability diagnosis and assessment model to obtain the corresponding instability assessment results. These instability assessment results characterize the changing trend of the coupler's operating state over time and the degree of its potential instability risk.

[0130] The coupler stability diagnosis results and the instability assessment results are output as the coupler instability diagnosis and assessment output results of the target locomotive, thereby completing the coupler instability diagnosis and assessment process based on multi-channel recognition results.

[0131] In one specific implementation, the final multi-channel identification model and diagnosis and evaluation model are used to diagnose and evaluate the instability state of the target heavy-duty locomotive coupler.

[0132] In this embodiment, the intelligent diagnosis and evaluation method for coupler instability of heavy-duty locomotives under limited measurement points described in this invention is used to perform longitudinal impulse multi-channel identification and locomotive operating status diagnosis and evaluation on the measured data. First, the multi-channel characterization identification is performed using the described heavy-duty locomotive longitudinal impulse multi-channel identification model, and the results are as follows: Figure 8 As shown. The results evaluation indices RMSE, MAE, and R were calculated. 2The values ​​are 24.3195, 15.4372, and 0.9871, respectively, indicating that the proposed method can achieve high-precision identification and can be used as a reliable technical means for multi-channel identification of longitudinal impulses in heavy-duty locomotives.

[0133] Secondly, based on the multi-channel identification results, the aforementioned coupler instability diagnosis and evaluation model was used to diagnose and evaluate the locomotive's operating status. The results show that at the current moment, the peak responses of the five channels—FCF, MCF, RCF, MCYA, and CM—did not exceed the specified thresholds, indicating that the locomotive is in a "stable" operating state. Simultaneously, the model predicts that the probabilities of each channel's response exceeding the set thresholds at the next moment are 36.44%, 47.45%, 61.21%, 16.50%, and 11.81%, respectively, indicating that the locomotive is likely to maintain stable operation in the near future.

[0134] In summary, this invention enables multi-channel indirect identification of longitudinal impulses in heavy-haul locomotives under limited measurement point conditions, and can simultaneously diagnose and assess the locomotive's operating status. Compared with existing technologies, this invention offers significant improvements in equipment cost, identification accuracy, operational reliability, and monitoring comprehensiveness.

[0135] In another possible implementation, during the long-term operation of heavy-haul locomotives, the operating sections corresponding to the train's traction and braking states are identified based on historical longitudinal impulse monitoring data, and the longitudinal impulse monitoring data within different operating sections are treated as independent analysis objects. For each operating section, filtering and noise reduction processing, multi-source channel correlation analysis, and limited measurement point channel screening processes are independently performed to obtain limited measurement point channel data matching the current operating section.

[0136] Based on this, a multi-channel identification model for longitudinal impulse of heavy-duty locomotives was constructed based on the limited measurement point channel data corresponding to different operating sections. The corresponding model was then used to perform multi-channel characterization and identification on the longitudinal impulse monitoring data in each operating section to obtain the section-related multi-channel identification results.

[0137] Furthermore, the multi-channel identification results obtained under different operating sections are input into the coupler instability diagnosis and evaluation model. By comparing and analyzing the coupler stability diagnosis results and instability evaluation results in different operating sections, the stability change characteristics of the coupler during the switching process between traction and braking states can be characterized. Through this implementation method, the coupler instability diagnosis and evaluation process can adapt to the differences in longitudinal impulse characteristics caused by changes in the operating conditions of heavy-haul locomotives.

[0138] In another possible implementation, during the initial operation of a heavy-haul locomotive or after maintenance, several historical longitudinal impulse monitoring data points with relatively stable operating conditions are selected. These data points are then processed using the method of this invention, including filtering and noise reduction, multi-source channel correlation analysis, and multi-channel characterization and identification, to generate multi-channel identification results for the corresponding time period. These multi-channel identification results are then stored as historical baseline states for the coupler's operational stability. During subsequent operation, the multi-channel characterization and identification process is repeated on longitudinal impulse monitoring data collected from the target locomotive under the same or similar operating conditions to obtain multi-channel identification results for the current operating state.

[0139] Furthermore, the current multi-channel identification results and the multi-channel identification results corresponding to the historical baseline state are jointly input into the coupler instability diagnosis and evaluation model. By analyzing the degree of deviation between the two in the multi-channel feature space, the diagnosis and evaluation of the changing trend of the coupler operating state are completed. Through this implementation method, the coupler instability diagnosis process can combine the locomotive's own historical operating characteristics to achieve state evaluation based on individual differences.

[0140] In another possible implementation, during the long-term operation of heavy-haul locomotives, some longitudinal impulse monitoring channels may experience data loss or decreased stability due to sensor aging, communication anomalies, or changes in installation status. To address this, when performing multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data, not only is limited measurement point channel data selected based on the correlation strength between channels, but the temporal continuity and data integrity of each monitoring channel are also constrained and judged simultaneously.

[0141] When an abnormal fluctuation or data interruption is detected in a monitoring channel within a preset time window, that monitoring channel is removed from the current set of limited measuring point channels, and the limited measuring point channel data is reconstructed based on the remaining monitoring channels. Subsequently, a multi-channel identification model for the longitudinal impulse of a heavy-haul locomotive is constructed using the updated limited measuring point channel data, and multi-channel characterization identification is performed on the longitudinal impulse state of the target locomotive at the current moment. Furthermore, the obtained multi-channel identification results are input into the coupler instability diagnosis and evaluation model to achieve coupler instability diagnosis and evaluation under conditions of partial measuring point degradation.

[0142] Figure 9 This is a system structure diagram of a locomotive coupler instability diagnosis and evaluation system provided in one embodiment of the present invention. Figure 9As shown, this invention provides a locomotive coupler instability diagnosis and assessment system. The system includes: a preprocessing unit for filtering and denoising historical longitudinal impulse monitoring data of existing heavy-duty locomotives to obtain filtered and denoised longitudinal impulse monitoring data; a processing unit for performing multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data; a feature recognition unit for constructing a multi-channel recognition model of the longitudinal impulse of heavy-duty locomotives based on the limited measurement point channel data, and using the multi-channel recognition model to perform multi-channel characterization and recognition of the limited measurement point channel data to obtain multi-channel recognition results; wherein, the multi-channel recognition model of the longitudinal impulse of heavy-duty locomotives is composed of a global-local multi-scale periodic-trend decomposition module and a multi-channel spatiotemporal integration fusion module connected in series; and an instability diagnosis unit for constructing a coupler instability diagnosis and assessment model based on the multi-channel recognition results, and diagnosing the coupler instability state of the target locomotive based on the coupler instability diagnosis and assessment model and outputting instability assessment results.

[0143] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described locomotive coupler instability diagnosis and assessment method.

[0144] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0145] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0146] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for diagnosing and assessing locomotive coupler instability, characterized in that, The method includes: The historical longitudinal impulse monitoring data of existing heavy-haul locomotives are filtered and denoised to obtain filtered and denoised longitudinal impulse monitoring data. Multi-source channel correlation analysis was performed on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data; Based on the limited measurement point channel data, a multi-channel recognition model for the longitudinal impulse of heavy-haul locomotives is constructed. This multi-channel recognition model is then used to perform multi-channel characterization and recognition on the limited measurement point channel data, yielding multi-channel recognition results. The multi-channel recognition model for longitudinal impulses of heavy-duty locomotives is composed of a global-local multi-scale period-trend decomposition module and a multi-channel spatiotemporal integral fusion module connected in series and working together. The method for constructing a multi-channel identification model for longitudinal impulses of heavy-duty locomotives based on the limited measurement point channel data includes: acquiring the limited measurement point channel data and using it as input data for the multi-channel identification model; constructing a global-local multi-scale periodic-trend decomposition module based on the limited measurement point channel data to decompose the data into multi-scale periodic and trend components, obtaining corresponding periodic and trend feature representations respectively; performing spatiotemporal dependency modeling on the multi-target monitoring channel data based on the periodic and trend feature representations, combining periodic-trend component modeling with explicit two-step numerical integration-driven data updates, and determining the modeling result as the output of the multi-channel spatiotemporal integration fusion module; and determining the output of the multi-channel spatiotemporal integration fusion module as the output of the multi-channel identification model for longitudinal impulses of heavy-duty locomotives to complete the construction of the multi-channel identification model for longitudinal impulses of heavy-duty locomotives. Based on the periodic feature representation and the trend feature representation, spatiotemporal dependency modeling is performed on the multi-target monitoring channel data through periodic-trend component modeling and combined with explicit two-step numerical integration-driven data updates. The modeling result is then determined as the output of the multi-channel spatiotemporal integration fusion module. This includes: constructing a periodic-trend component modeling hybrid submodule of the multi-channel spatiotemporal integration fusion module based on the periodic feature representation and the trend feature representation; and performing component modeling processing to weaken nonlinear interference on the periodic feature representation and the trend feature representation through the periodic-trend component modeling hybrid submodule to obtain the periodic-trend component modeling hybrid... The modeling results of the multi-channel spatiotemporal integration module are as follows: Based on the initial target monitoring channel identification results, a reversible normalized Fourier plot submodule of the multi-channel spatiotemporal integration fusion module is constructed by driving data updates through explicit two-step numerical integration. Non-stationarity suppression and spatiotemporal dependency modeling are performed on the multi-target monitoring channel data through the reversible normalized Fourier plot submodule to obtain the modeling results of the reversible normalized Fourier plot submodule. The modeling results of the periodic-trend component modeling hybrid submodule and the modeling results of the reversible normalized Fourier plot submodule are summed, and the summation result is determined as the output result of the multi-channel spatiotemporal integration fusion module. Based on the multi-channel recognition results, a coupler instability diagnosis and evaluation model is constructed, and the coupler instability state of the target locomotive is diagnosed and the instability evaluation results are output based on the coupler instability diagnosis and evaluation model.

2. The locomotive coupler instability diagnosis and assessment method according to claim 1, characterized in that, Historical longitudinal impulse monitoring data from existing heavy-haul locomotives are filtered and denoised to obtain filtered and denoised longitudinal impulse monitoring data, including: The historical longitudinal impulse monitoring data is divided into coupler force monitoring data and displacement monitoring data according to the type of monitored physical quantity. Based on the spectral characteristics of the coupler force monitoring data and the displacement monitoring data, corresponding low-pass filter cutoff frequencies are set for the coupler force monitoring data and the displacement monitoring data, respectively. The coupler force monitoring data is subjected to low-pass filtering processing using a low-pass filter cutoff frequency corresponding to the coupler force monitoring data. The displacement monitoring data is subjected to low-pass filtering processing using a low-pass filter cutoff frequency corresponding to the displacement monitoring data. The filtered coupler force monitoring data and the filtered displacement monitoring data are time-aligned and integrated to form filtered and noise-reduced longitudinal impulse monitoring data.

3. The locomotive coupler instability diagnosis and assessment method according to claim 1, characterized in that, Multi-source channel correlation analysis was performed on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data, including: The filtered and denoised longitudinal impulse monitoring data is divided according to the monitoring channels to form multiple longitudinal impulse monitoring channel data sequences; Based on the longitudinal impulse monitoring channel data sequence, the correlation coefficient between any two monitoring channels is calculated using the Pearson correlation coefficient to obtain the correlation distribution results between each monitoring channel. Based on the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the longitudinal impulse level of the locomotive is calculated respectively. Based on the correlation strength, monitoring channels that meet preset conditions are selected from multiple longitudinal impulse monitoring channels to provide limited measurement point channel data.

4. The locomotive coupler instability diagnosis and assessment method according to claim 3, characterized in that, Based on the correlation distribution results, the correlation strength between each monitoring channel and the target monitoring channel used to characterize the locomotive's longitudinal impulse level is calculated, including: In the correlation distribution results, a correlation coefficient sequence corresponding to the target monitoring channel is selected. The correlation coefficient sequence consists of the correlation coefficients between the target monitoring channel and the other monitoring channels. Based on the correlation coefficient sequence, the correlation strength of each monitoring channel relative to the target monitoring channel is determined to characterize the response of the corresponding monitoring channel to changes in the longitudinal impulse level of the locomotive.

5. The locomotive coupler instability diagnosis and assessment method according to claim 1, characterized in that, Based on the multi-channel recognition results, a coupler instability diagnosis and evaluation model is constructed, including: The multi-channel recognition results are obtained and used as input data for the coupler instability diagnosis and evaluation model. Based on the multi-channel feature information reflecting the changes in the coupler's operating state in the multi-channel recognition results, a state representation structure for characterizing the coupler's operating stability is constructed. Based on the state representation structure, a diagnostic discrimination structure is constructed to distinguish different stable operating states of couplers; The diagnostic discrimination structure is combined with the assessment structure used to characterize the degree of instability risk to form a coupler instability diagnosis and assessment model.

6. The locomotive coupler instability diagnosis and assessment method according to claim 5, characterized in that, Based on the aforementioned coupler instability diagnosis and evaluation model, the coupler instability state of the target locomotive is diagnosed and the instability evaluation results are output, including: The multi-channel identification results of the target locomotive at the corresponding time are input into the coupler instability diagnosis and evaluation model; Based on the coupler instability diagnosis and evaluation model, stability discrimination calculation is performed on the multi-channel identification results to obtain the coupler stability diagnosis result of the target locomotive at the current moment. Based on the obtained coupler stability diagnosis results, the time evolution trend of the coupler operating state is evaluated and calculated based on the coupler instability diagnosis and evaluation model to obtain the corresponding instability evaluation results. The coupler stability diagnosis results and the instability assessment results are used as the coupler instability diagnosis and assessment output results for the target locomotive.

7. A locomotive coupler instability diagnosis and assessment system, characterized in that, The system is used to perform the locomotive coupler instability diagnosis and assessment method according to any one of claims 1-6, the system comprising: The preprocessing unit is used to filter and reduce noise on the historical longitudinal impulse monitoring data of existing heavy-duty locomotives to obtain filtered and denoised longitudinal impulse monitoring data. The processing unit is used to perform multi-source channel correlation analysis on the filtered and denoised longitudinal impulse monitoring data to obtain limited measurement point channel data; The feature recognition unit is used to construct a multi-channel recognition model for the longitudinal impulse of a heavy-haul locomotive based on the limited measurement point channel data, and to use the multi-channel recognition model to perform multi-channel characterization and recognition on the limited measurement point channel data to obtain multi-channel recognition results; wherein, The multi-channel recognition model for longitudinal impulses of heavy-duty locomotives is composed of a global-local multi-scale period-trend decomposition module and a multi-channel spatiotemporal integral fusion module connected in series and working together. The instability diagnosis unit is used to construct a coupler instability diagnosis and evaluation model based on the multi-channel recognition results, and to diagnose the coupler instability state of the target locomotive based on the coupler instability diagnosis and evaluation model and output the instability evaluation results.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the locomotive coupler instability diagnosis and assessment method as described in any one of claims 1-6.

Images