Railway intelligent operation management evaluation method and system

Abstract

The invention relates to the technical field of railway operation management, and discloses a railway intelligent operation management evaluation method and system. The method comprises the following steps: acquiring historical railway operation data, extracting a train operation index, a track state index and an environment interference index, and constructing an operation evaluation model according to the indexes; setting an analysis time interval, performing sliding segmentation on the orbit state indexes based on the interval to obtain an orbit state data subset, and analyzing the subset to extract orbit state fluctuation features and anomaly occurrence features; and obtaining a current evaluation moment, matching a corresponding track state index according to the track state fluctuation characteristics, calculating the abnormity possibility of the index in combination with the abnormity occurrence characteristics, and outputting the track state index and the abnormity possibility thereof. According to the method, multi-dimensional data are integrated, the track state change is scientifically analyzed, the abnormal possibility is accurately judged, the comprehensiveness, scientificity and accuracy of railway operation management evaluation are improved, and the modern railway intelligent operation management requirement is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway operation management, in particular to a railway intelligent operation management evaluation method and system. BACKGROUND

[0002] In the railway transportation industry, operation management evaluation is an important work content to ensure the safe and stable operation of railways. With the continuous expansion of the railway network scale, the continuous increase of train operation frequency, and the increasing influence of complex natural environment on railway operation, the traditional railway operation management evaluation method gradually shows many limitations.

[0003] The traditional evaluation method relies on manual data collection, and the timeliness and completeness of data acquisition cannot be guaranteed. The staff usually needs to detect the track state and record the train operation parameters regularly. This method not only consumes a lot of manpower and material resources, but also may cause the accuracy of data to decrease due to human operation errors. At the same time, the traditional method is simple in data processing, and usually only analyzes a single index independently, ignoring the correlation between train operation index, track state index and environmental interference index, and cannot form a comprehensive operation evaluation system.

[0004] In terms of track state analysis, the traditional method lacks effective time dimension processing means. In most cases, only the track data at a certain time point is evaluated, which cannot reflect the law of track state change over time and is difficult to identify potential abnormalities of track state in advance. For example, when the track appears slight wear or deformation, the data at a single time point may not be able to capture these subtle changes in time, and these changes may cause serious safety accidents if accumulated for a long time.

[0005] In addition, the traditional evaluation method lacks scientific basis in abnormality possibility judgment. Usually, the staff's experience is relied on to judge whether the track state is abnormal, and this subjective judgment method is greatly influenced by individual experience level, and is easy to cause misjudgment or omission. For the changes of track state under different environmental conditions, the traditional method cannot establish an effective correlation analysis model, which reduces the reliability and practicality of the evaluation results, and is difficult to meet the needs of modern railway intelligent operation management.

[0006] With the acceleration of the development process of railway intelligence, higher requirements are put forward for the accuracy, real-time and comprehensiveness of operation management evaluation. The traditional evaluation method cannot meet the needs of railway operation management under the new situation, and an intelligent evaluation method is needed which can integrate multi-dimensional data, scientifically analyze the changes of track state, and accurately judge the possibility of abnormality. SUMMARY

[0007] The present application aims to provide a railway intelligent operation management evaluation method and system to solve the problems raised in the background.

[0008] To achieve the above object, the present application provides a railway intelligent operation management evaluation method, which comprises: acquiring railway historical operation data, extracting train operation indexes, track state indexes and environmental interference indexes, and constructing an operation evaluation model according to the train operation indexes, the track state indexes and the environmental interference indexes; setting an analysis time interval, performing sliding segmentation on the track state indexes based on the analysis time interval to obtain a track state data subset, analyzing the track state data subset, and extracting track state fluctuation characteristics and abnormal occurrence characteristics; acquiring a current evaluation time, matching corresponding track state indexes according to the track state fluctuation characteristics, combining the abnormal occurrence characteristics to calculate the abnormal possibility of the track state indexes, and outputting the track state indexes and the abnormal possibility thereof.

[0009] Preferably, the train operation indexes include train running speed, train load distribution and train braking performance, the track state indexes include track wear degree, track deformation and track temperature gradient, and the environmental interference indexes include wind speed change, humidity change and temperature change. The calculation method of the track wear degree comprises: calculating the difference value between the actual wear amount of the track and the standard wear amount, denoted as wear difference value, calculating the proportion of the wear difference value to the standard wear amount, and setting the proportion as the track wear degree. The calculation method of the track deformation comprises: acquiring track deformation monitoring data, calculating the deformation change amount within a unit time, and setting the deformation change amount as the track deformation.

[0010] Preferably, the train operation indexes are set as input elements, the track state indexes and the environmental interference indexes are set as interference elements, and the train operation safety evaluation value is set as an output element. The construction method of the operation evaluation model comprises: performing time-frequency conversion on the input elements and the interference elements, constructing an operation transfer function based on the converted input elements and output elements, introducing the interference elements, and converting the operation transfer function into an operation evaluation model.

[0011] Preferably, a first preset time length is set as the analysis time interval length, and a second preset time length is set as the moving step. The track state indexes are arranged into a time series data sequence, based on the sliding window technology, the window is moved according to the moving step from the starting point of the time series data sequence, and the track state data subset in each window is extracted.

[0012] Preferably, spectral feature analysis is performed on each track state data subset, the power density distribution of the track state indicator is calculated, the power density distribution interval higher than the preset threshold is selected, the time interval average of adjacent intervals is calculated, the time interval average is set as the track state fluctuation feature, and the correspondence between the track state indicator and the fluctuation feature is established.

[0013] Preferably, the total number of track state data subsets is counted, the abnormal occurrence frequency of each track state indicator in the data subset is counted, the ratio of the abnormal occurrence frequency to the total number of data subsets is calculated, the ratio is set as the abnormal occurrence feature, and the correspondence between the track state indicator and the abnormal occurrence feature is established.

[0014] Preferably, the current evaluation time is obtained, the time difference between the current time and the end point of the previous track state data subset is calculated, and is recorded as an evaluation time difference; The deviation amount of the evaluation time difference and the fluctuation feature corresponding to each track state indicator is calculated, and the track state indicator corresponding to the minimum deviation amount is selected as the predicted track state indicator; Based on the correspondence between the track state indicator and the abnormal occurrence feature, the abnormal possibility of the predicted track state indicator is matched.

[0015] Preferably, according to the predicted track state indicator and the abnormal possibility thereof, a high-risk track section is screened, train operation indicators and environmental interference indicators of the high-risk track section are extracted, and a train operation safety evaluation value is calculated in combination with an operation evaluation model.

[0016] Preferably, based on the train operation safety evaluation value, a railway operation optimization scheme is generated, and a train operation scheduling scheme or a track maintenance plan scheme is adjusted.

[0017] Preferably, the present application also includes a railway intelligent operation management evaluation system, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor realizes the steps of the railway intelligent operation management evaluation method when executing the computer program.

[0018] Compared with the prior art, the present application has the following advantages: The railway intelligent operation management evaluation method integrates multi-dimensional data to construct an operation evaluation model, and realizes comprehensive evaluation of the railway operation state. It can simultaneously obtain train operation indicators, track state indicators and environmental interference indicators in the railway historical operation data, and construct an operation evaluation model based on these indicators, breaking the limitation of single indicator analysis in traditional methods, fully considering the internal correlation between indicators, so that the evaluation model can more comprehensively reflect the overall condition of the railway operation.

[0019] In the track state analysis link, the method sets the analysis time interval and performs sliding segmentation on the track state indicators, obtains the track state data subset, and then extracts the track state fluctuation characteristics and abnormal occurrence characteristics. This sliding segmentation processing method based on the time dimension can effectively capture the law of the change of the track state with time, compared with the traditional method of evaluating only the data at a specific time point, it is more conducive to discovering the change trend of the track state in different time periods and timely identifying the subtle fluctuations of the track state, providing more abundant basis for subsequent abnormality possibility judgment.

[0020] In the abnormality possibility judgment aspect, the method combines the current evaluation time, matches the corresponding track state indicators according to the track state fluctuation characteristics, and calculates the abnormality possibility in combination with the abnormal occurrence characteristics. This feature matching-based abnormality judgment method gets rid of the subjective limitations of the traditional method relying on manual experience judgment, and obtains the abnormality possibility result through objective feature analysis and data calculation, making the judgment process more scientific and accurate. It can avoid misjudgment or omission caused by individual experience differences, and ensure that the judgment of track state abnormality is more reliable.

[0021] The method can output the track state indicators and their abnormality possibility, providing intuitive and effective evaluation results for railway operation management. Related personnel can clearly understand the current state of the railway track and possible abnormal conditions according to these evaluation results, so as to carry out targeted operation management. Whether in the daily railway maintenance plan formulation or in the decision-making in emergency situations, the evaluation results can play an important role, helping to improve the efficiency and quality of railway operation management, promoting the intelligent and precise development of railway operation management, and better adapting to the needs of modern large-scale and high-frequency railway operation.

[0022] The implementation process of the method does not require complex hardware device support and can be completed based on existing data acquisition and processing technology, which has strong practicality and operability. It can improve the operation management evaluation level without significantly increasing the railway operation cost, providing a strong guarantee for the sustainable development of the railway industry. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The working principle diagram of the railway intelligent operation management evaluation method described in the present application; Figure 2 The working principle diagram of the index definition and calculation of the railway intelligent operation management evaluation method; Figure 3 The working principle diagram of the operation evaluation model construction of the railway intelligent operation management evaluation method. DETAILED DESCRIPTION

[0024] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0025] Please refer to Figure 1 The present application provides a railway intelligent operation management evaluation method and system. The method includes dynamic evaluation of operation status by integrating historical data analysis and real-time monitoring functions. The method obtains railway historical operation data from train operation record systems, track monitoring equipment, and environmental sensors. Train operation indicators, track status indicators, and environmental interference indicators are extracted from the historical data. Train operation indicators reflect train operation parameters, track status indicators describe track physical conditions, and environmental interference indicators capture external influencing factors. An operation evaluation model is constructed based on these indicators, which can simulate the interaction relationships in railway operation. An analysis time interval is set, which is a continuous time range used to focus on the data analysis period. The track status indicators are divided into subsets using a time window sliding method based on the analysis time interval. Each track status data subset is analyzed to extract track status fluctuation characteristics and abnormal occurrence characteristics. Fluctuation characteristics represent periodic or trend changes in track status, and abnormal occurrence characteristics quantify the frequency of abnormal occurrences. The current evaluation time, i.e., the specific time point for evaluation, is obtained. The corresponding track status indicators are matched based on the track status fluctuation characteristics, and the matching process is achieved by comparing the similarity between the current time and historical fluctuation patterns. The abnormal probability of the track status indicators is calculated based on the abnormal occurrence characteristics, and the abnormal probability is output in the form of a probability value.

[0026] Example 1: Please refer to Figure 2Train operation indicators include running speed, load distribution, and braking performance. Running speed data is collected by speed sensors installed on train axles or drive systems, outputting high-frequency pulses or analog signals, which are converted by on-board data recording units into digital time series. Load distribution relies on an array of weight sensors at the bottom of each carriage, measuring each support pressure and calculating total weight and center of gravity. Data is transmitted wirelessly or aggregated through the vehicle network. Braking performance is evaluated based on parameters recorded by the brake system controller, such as brake cylinder pressure. These indicators form the basic data set for train operation status. Track status indicators include wear level, deformation, and temperature gradient. Wear level is obtained through regular inspections or fixed measurement equipment. Environmental interference indicators include wind speed, humidity, and temperature changes. Wind speed data is collected from ultrasonic anemometers at weather stations along the railway at a frequency of seconds. Humidity is monitored by capacitive or resistive humidity sensors at trackside shelters. Temperature is integrated from weather station and track temperature sensor data, calculating daily or hourly change rates to reflect environmental thermal load.

[0027] Track wear level calculation: Determine the standard wear amount based on track material, designed traffic volume, and maintenance specifications, obtained through historical statistics or laboratory tests, which may be dynamically updated. Measure the actual wear amount on site, calculate the difference from the standard wear amount, and then calculate the wear level value by proportion, normalized. The standard wear amount may be customized by section in the calculation, dynamically updated based on recent maintenance or material replacement. The measurement device needs to be aligned with the coordinates, and after calculation, detect anomalies and remove outliers. The proportion calculation introduces logarithmic transformation. If a distributed measurement network is used, data synchronization relies on a time server, and the wear difference value is averaged using a sliding window. The result is output in levels. Track deformation calculation: Obtain monitoring data from displacement sensors or total station instrument measurement systems at key points on the track, calculate the deformation difference between the current and previous periods, and generate a line density indicator by normalization. If the monitoring data comes from an automated system, it needs to be processed for transmission delay and packet loss, and missing values need to be completed. The unit time selection considers the deformation rate. The calculation distinguishes between elastic and plastic deformation. If there is baseline drift in the displacement sensor data, it needs to be corrected, and relative value calculation can reduce installation errors. The unit time is synchronized with train passing frequency.

[0028] Track temperature gradient calculation: integrate multi-point temperature data, calculate gradient with difference method, environmental interference index processing wind speed, humidity, temperature change, all through data cleaning and smoothing processing. Wind speed change analysis can add wind direction, turbulence intensity, gust factor, altitude correction; humidity change monitoring avoids condensation, direct sunlight, considers precipitation events; temperature change data is linked with track temperature gradient, seasonal pattern, track stress model. Train running speed data processing: filter and calibrate the original pulse signal, aggregate into average value. Train load distribution calculation: fuse weight sensor data, temperature compensation and zero drift correction, synthesize each car load, generate load distribution curve. Train braking performance evaluation: integrate dynamic braking test data, associate operating conditions to establish performance degradation model. Data processing, all index data storage time stamp and location label, calculation process embedded in data pipeline, realize real-time or near real-time processing. Train running index data source is extensible, track state index calculation can integrate non-destructive testing data. The whole implementation focuses on data quality control, consistency check after index calculation, optimization of calculation efficiency, cache results support fast query.

[0029] Train running speed data processing adds running condition classification, such as traction, cruise, braking phase analysis respectively, load distribution calculation considers the difference of cargo types, braking performance evaluation adds slope and curve correction. Track wear degree calculation using machine learning method, train wear prediction model, actual wear input model output degree value, deformation variable analysis adds frequency spectrum analysis to detect resonance frequency. Wind speed change calculation of environmental index calculates turbulence dissipation rate, humidity change monitoring combines with evaporation heat dissipation model, temperature change analysis considers the influence of daily range on track stability. All indicators record metadata such as sensor ID and calibration date in implementation, support traceability audit. Track deformation variable calculation specific steps may include coordinate transformation of original displacement data, from sensor local system to track global system, deformation change calculation uses second difference to capture acceleration change, unit time setting based on track structure natural frequency. Data preprocessing of environmental interference index includes outlier rejection, using box plot or Z-score method, wind speed change calculation adds wind rose diagram analysis of dominant wind direction, humidity change monitoring avoids freezing misjudgment, temperature change data is associated with solar radiation intensity. Index calculation is stored in time series database, supports time range query and aggregation operation.

[0030] In the train operation index, the speed data may compensate for the wheel diameter wear error, the load distribution calculation adds a dynamic allocation algorithm, and the brake performance evaluation integrates the friction coefficient estimation. If the track wear degree calculation uses contact type measurement, the probe needs to be calibrated regularly, and the deformation variable analysis adds track geometric parameters such as track gauge and level. The wind speed change calculation of the environmental index calculates the Beaufort level, the humidity change monitoring combines with the corrosion risk evaluation, and the temperature change analysis analyzes the thermal stress accumulation. The entire implementation is embedded in a quality assurance cycle, and the index calculation results are fed back to the measurement system to adjust the sampling strategy, forming a closed-loop optimization. In the track wear degree calculation, the standard wear amount may be set differently according to the track type such as high-speed line or freight line, the actual wear amount measurement needs image stitching and three-dimensional reconstruction if using unmanned aerial vehicle inspection, and the wear difference value calculation uses statistical hypothesis testing to identify significant changes, and the proportion calculation is mapped to the maintenance priority index. In the track deformation variable calculation, the data fusion of the displacement sensor and the inertial measurement unit improves the accuracy, the deformation change calculation adds a time series prediction model, and the unit time adjustment is based on the deformation rate adaptability. The wind speed change analysis of the environmental index adds a terrain amplification factor, the humidity change monitoring avoids sensor pollution, and the temperature change data is combined with the material thermal expansion coefficient. The index calculation process is visualized and monitored, and the abnormal calculation triggers an alarm, and manual review is involved.

[0031] The data integration of the train operation index comes from multiple heterogeneous systems, the speed data aligns the time stamp to solve the clock drift, the load distribution calculation adds vibration compensation, and the brake performance evaluation considers the regenerative braking contribution. The track state index calculation introduces digital twin technology, the wear degree simulates the material wear mechanism, and the deformation variable analysis couples the vehicle-track dynamics model. The environmental disturbance index is extended to seismic vibration monitoring, the wind speed change calculation calculates the wind load force, the humidity change analysis analyzes the condensation risk, and the temperature change prediction predicts the steel rail temperature field. In the implementation, attention is paid to scalability, the index calculation is modularly designed, and new index plug-in is supported. In the in-depth level of track deformation variable calculation, it may involve distributed fiber sensing technology, continuous spatial sampling of deformation data, deformation change calculation using strain rate instead of displacement, and unit time synchronization with train passing events to capture instantaneous deformation. The wind speed change analysis adds Computational Fluid Dynamics simulation verification, the humidity change monitoring combines with the electrolyte sensor to improve the accuracy, and the temperature change data is integrated with the climate change model. All index calculations follow the data minimization principle, only the feature values are retained to reduce the storage load, and the calculation process log is recorded for performance optimization.

[0032] Example 2: see Figure 3, train operation indexes are set as input elements, which represent the core parameters that can be actively regulated in operation management, and these elements are directly controlled by dispatching instructions. Track state indexes and environmental disturbance indexes are set as interference elements, which reflect the variables that are difficult to control in real time from the outside or inside of the system. Track state indexes such as wear degree and deformation reflect the inherent conditions of infrastructure, and environmental disturbance indexes such as wind speed and temperature changes capture the influence of external natural forces that cannot be resisted. Train operation safety evaluation value is explicitly defined as an output element, which is a comprehensive quantitative value to represent the overall safety level of railway operation under given input and interference conditions, and its value is directly related to the risk level. The core of the construction method of the operation evaluation model is to convert time domain data into frequency domain features to reveal potential laws. The input elements and interference elements are first subjected to time-frequency conversion, which is usually realized by using fast Fourier transform algorithm or wavelet transform algorithm. The conversion process decomposes time series data into components of different frequencies, highlighting the periodicity, trend and burstiness of the data. The operation transfer function is constructed based on the converted input elements and output elements. The construction work relies on the historical data set, and the mathematical relationship between input and output is found through system identification method. The introduction of interference elements is the key step to transform the initially established operation transfer function into a more practical operation evaluation model. The interference elements are integrated into the transfer function framework in the form of additional terms or modulation factors. The specific operation of time-frequency conversion needs to deal with data preprocessing and parameter selection problems. The original data needs to be detrended and standardized before conversion to eliminate dimension difference and baseline drift. The window length of Fourier transform or the mother wavelet type of wavelet transform is selected according to the characteristics of the data to balance the frequency resolution and time resolution.

[0033] The construction of the operation transfer function relies on a large number of historical operation cases, each of which contains a specific combination of input elements, interference element state, and actual recorded safety evaluation results, which may come from post-analysis reports or real-time safety monitoring system judgment values. The function fitting process may use a recursive algorithm to gradually optimize the parameters, so that the error between the output value of the model and the historical true value is minimized. The way of introducing interference elements needs to consider their interaction with input elements, sometimes an interference element impact model on transfer function parameters is established instead of simple superposition. Data synchronization is the basis of model construction, the data of input elements, interference elements and output elements must be strictly aligned in time and space position, for different sampling frequency data streams need to be resampled or interpolated to achieve time alignment. The feature vector after time-frequency conversion has high dimension, principal component analysis or other methods can be used for dimension reduction to retain the main frequency components to reduce the computational complexity and prevent overfitting. The structure of the operation transfer function may choose a linear transfer function or a nonlinear neural network structure, which depends on the complexity of the system dynamic characteristics, linear models are easy to analyze and understand, while nonlinear models may have stronger fitting ability.

[0034] The weight distribution of interference elements is not fixed, it can be trained and learned based on historical data by machine learning algorithms, so that the model can adaptively adjust the importance of different interference factors. In the model training stage, the data set is divided into training set, validation set and test set, the training set is used to preliminarily determine the transfer function parameters, the validation set is used to adjust the model hyperparameters and prevent overfitting, and the test set is used to finally evaluate the generalization performance of the model. The model output, i.e. the train operation safety evaluation value, is usually standardized to a fixed numerical interval, such as between 0 and 1, to facilitate unified interpretation and setting of warning thresholds. The model updating mechanism needs to consider the concept drift problem, i.e. the slow change of system characteristics over time, so it needs to retrain or update the model parameters online regularly to maintain its prediction accuracy. The whole construction process embeds version control and quality check links, each model update records its configuration parameters and performance indicators, ensuring the traceability of model changes. The specific form of the operation transfer function may be a multi-input single-output difference equation or a state space equation, and its parameter estimation involves complex numerical optimization algorithms. The introduction of interference elements may use a product model or an additive model, the product model considers that interference will amplify or reduce the influence of input, while the additive model considers that interference produces an independent additional effect. Model validation not only focuses on overall accuracy but also on its performance in extreme conditions, in practice, a variety of methods may be used according to the characteristics of the index data.

[0035] The establishment of the operation transfer function can be done in stages, first establishing a baseline model without considering interference and then gradually introducing interference factors for correction. This progressive approach helps to understand the contribution of each factor. The data of interference factors may have missing values or high noise, and the model needs to have certain robustness, such as handling missing values through the expectation maximization algorithm or using robust regression methods to resist abnormal value interference. The interpretability of the model is particularly important in safety-critical areas, so even if a black box model such as a deep learning network is used, it needs to be supplemented with SHAP or LIME interpretation tools to understand the basis for model decision-making. The final operation evaluation model is a software module that encapsulates all the calculation logic and parameters, providing a standardized application programming interface for other systems to call. The model input interface receives real-time data streams or batch data files, and outputs safety evaluation values and possible confidence intervals after internal processing. Model performance monitoring is ongoing, and once a persistent increase in prediction bias is detected, an alarm is triggered and a model review is prompted. During the construction process, domain knowledge is used to guide feature engineering and model structure design, and the entire model construction life cycle management follows a strict process, from requirement analysis, data preparation, feature engineering, model training, verification and testing to deployment and operation monitoring. Each stage has corresponding document records and approval nodes. The operation evaluation model ultimately realizes a mapping function that maps high-dimensional, dynamic input and interference spaces to a one-dimensional safety evaluation value. This mapping relationship captures the inherent physical and statistical laws of the railway operation system. The application scenarios of the model include real-time safety monitoring, operation scheme simulation testing, and maintenance plan optimization, etc. It provides a quantitative, data-based reference for railway managers' decision-making. The long-term effectiveness of the model depends on a continuous learning cycle, with new data generated during operation continuously fed back to the model for parameter tuning and structure evolution, enabling it to adapt to the continuous changes in the railway system and external environment.

[0036] Embodiment 3: The first preset duration is endowed with the role of analysis time interval length, which is a fixed time span, the specific value of which is determined according to the continuity of data collection, the urgency of track state change and the depth of retrospective required by the evaluation task. The second preset duration is defined as the moving step, which determines the distance of the sliding window sliding forward each time, and its value is usually less than or equal to the first preset duration, such as being set to one day or one week, to ensure that there is both overlap and progress between adjacent data subsets. Arranging track state indicators into time series data sequences is the basis for subsequent processing, which requires cleaning, aligning and time sorting of raw track state data from different monitoring points and different sensors to form a strictly increasing data sequence according to time stamp, and each data point in the sequence is associated with a specific track section and collection time. Based on the sliding window technology, the window is moved according to the moving step from the starting point of the time series data sequence, the initial position of the window covers the first segment of data in the sequence, and the length of the window is equal to the first preset duration. Each sliding operation makes the window move forward along the time axis by a distance of a moving step, thereby generating a new data coverage interval. Extract the track state data subsets in each window, each data subset is essentially a continuous slice of the original time series sequence, containing the specific values of all track state indicators within the window time range, and these subsets constitute independent samples for subsequent feature analysis.

[0037] The spectral feature analysis of each track state data subset is the core step of extracting fluctuation characteristics. The spectral analysis uses the method of digital signal processing, and each data subset is regarded as a discrete time signal. The purpose is to transform the signal from the time domain to the frequency domain to reveal its inherent frequency components. The power density distribution of the track state index is calculated, which reflects the distribution of signal power at different frequency points. Its calculation is usually based on the periodogram method or the more advanced Welch method, which reduces variance through segmentation averaging to obtain a smooth curve showing the relationship between power and frequency. The power density distribution interval higher than the preset threshold is selected. The preset threshold is a manually set threshold value used to filter out background noise and secondary fluctuations, and only the main frequency components with significant energy are retained. These selected intervals correspond to the periodic fluctuations in the track state. The time interval average of adjacent intervals is calculated. Here, "adjacent intervals" refer to the frequency intervals corresponding to the selected peak values higher than the threshold on the power density distribution curve. The time interval (i.e., the inverse of the period) between these peak frequencies is calculated, and then the arithmetic mean of all such interval values is calculated. The time interval average is set as the track state fluctuation characteristic. This characteristic value quantifies the average time scale of the main periodic fluctuations of the track state. The corresponding relationship between the track state index and the fluctuation characteristic is established. This relationship is stored in the form of a data mapping table or an associated database, recording the fluctuation characteristic value corresponding to each track state data subset, thereby forming a knowledge base of historical fluctuation patterns.

[0038] The specific implementation of the sliding window technique needs to consider the sequence boundary conditions. When the window slides to the end of the sequence, its coverage may be insufficient for the first preset time length. At this time, zero padding or truncation processing can be used. The extraction of data subsets needs to ensure time continuity and avoid the occurrence of holes in the window due to data loss, which usually requires interpolation or marking processing. Before spectral analysis, de-meaning and windowing of each data subset are standard procedures, such as using the Hanning window to reduce spectral leakage. The calculation of the power density distribution involves the fast Fourier transform, and its frequency resolution is directly related to the window length. The selection of the first preset time length affects the minimum frequency interval that can be distinguished.

[0039] The preset threshold is not fixed; it can be adaptively adjusted based on the overall signal noise level, for example, set to a multiple of the full-spectrum power mean. After selecting an interval above the threshold, peak detection is sometimes necessary to accurately locate the dominant frequency. When calculating the time interval between adjacent peaks, the frequency axis order must be considered to ensure that the interval calculation reflects the periodic difference between consecutive dominant frequencies. The storage of fluctuation characteristics includes not only the mean but also sometimes the standard deviation or other statistics to describe the stability of the characteristics. The entire processing flow places certain demands on computing resources, especially when the time series is very long, requiring algorithm optimization or distributed computing. Track condition indicators may include multiple variables, such as wear degree and deformation. Spectrum analysis can be performed individually for each variable or jointly analyzed on multivariate signals to examine their coherence. The calculated power density distribution is a discrete frequency function, requiring the search for its local maxima to determine the dominant frequency interval. The selection of the preset threshold is somewhat subjective and needs to be determined through historical data analysis or experience to find a value that can effectively distinguish between real fluctuations and noise. When establishing the correspondence, in addition to the fluctuation characteristic value, the timestamp of the data subset that generated the characteristic and the track segment identifier are also recorded for subsequent traceability.

[0040] In engineering practice, time-series data sequences may contain missing values ​​due to sensor failures or communication interruptions. Data imputation or labeling is necessary before constructing the sequence. Spectral analysis is sensitive to data continuity; improper handling of missing values ​​can introduce spurious frequency components. The choice of the sliding window's step size is a trade-off: smaller step sizes generate more data subsets, which is beneficial for capturing subtle changes in features but increases computation; larger step sizes are more computationally efficient but may miss some transient events. The estimation method of power density distribution affects the accuracy of features. The Welch method increases the sample size through overlapping segments, resulting in a smaller variance in the spectral estimate, but a slight decrease in frequency resolution.

[0041] The formula for calculating the fluctuation characteristics can be expressed as:

[0042] Where: symbol This represents the average time interval obtained at the end, i.e., the orbital state fluctuation characteristic. (Symbol) This represents the total number of dominant frequency ranges identified in the power density distribution that are above a preset threshold. (Symbol) and These represent the center frequency values ​​of the i-th and (i+1)-th selected dominant frequency intervals, respectively. (Symbol) This represents the summation of all valid adjacent intervals. This formula quantifies the average time span between major fluctuation patterns.

[0043] The stability of the fluctuation features needs to be monitored. If the computed mean time interval varies dramatically between different data subsets, it may indicate that the track condition is unstable or there is an unidentified disturbance factor. The established correspondence database needs to be updated periodically to reflect the evolving trend of the track condition over time, and old, unrepresentative feature mappings need to be archived or discarded. The entire implementation process is embedded in automated scripts, from data preparation, window sliding, spectrum analysis to feature calculation and storage, to minimize human intervention and improve processing efficiency and consistency. The resulting fluctuation features and their correspondence with the track condition indicators provide a historical benchmark for identifying the fluctuation pattern to which the current state belongs in the subsequent evaluation phase.

[0044] In Example 4, the total number of track condition data subsets is the starting point of this work, which is directly determined by the sliding window segmentation process. The abnormal occurrence frequency of each track condition indicator in the data subset is counted, which requires first determining the normal value range of each track condition indicator. For example, the track wear degree may be set with an upper threshold value, and values exceeding the threshold value are considered abnormal. The statistical operation traverses each data subset to check whether the value of a specific track condition indicator exceeds its threshold value, and accumulates the number of abnormalities that occur in all data subsets.

[0045] The ratio of the abnormal occurrence frequency to the total number of data subsets is calculated, which is a simple division operation. The ratio is set as the abnormal occurrence feature, which is a probability estimate between 0 and 1, indicating the likelihood of the occurrence of an abnormality in a single analysis time interval based on historical data. The correspondence between the track condition indicators and the abnormal occurrence features is established, which is usually maintained in the form of a mapping table or database index, so that given a track condition indicator, its corresponding historical abnormal occurrence feature value can be immediately queried. The current evaluation time is the trigger point for real-time evaluation using these historical features. The current evaluation time may be triggered by a system timing task or manually specified by a user. The time difference between the current time and the end point of the previous track condition data subset is calculated. Assuming that the data subsets are processed in chronological order, the end point of the previous subset is the end time stamp of the most recently analyzed window. This time difference is denoted as the evaluation time difference, which reflects the time span from the current time to the most recent complete historical analysis. The deviation of the evaluation time difference from the fluctuation feature corresponding to each track condition indicator is calculated. The fluctuation feature represents the typical time interval of the historical fluctuation of the indicator, and the deviation is the absolute difference between the evaluation time difference and this typical time interval. This calculation is performed for each track condition indicator with an established fluctuation feature. The track condition indicator corresponding to the smallest deviation is selected as the predicted track condition indicator. The smallest deviation means that the current time is closest to a certain historical fluctuation pattern, so it is reasonable to predict that the current track condition is most likely to follow the fluctuation pattern represented by that specific indicator.

[0046] Based on the correspondence between the track state indicators and the abnormal occurrence characteristics, the abnormal possibility of the predicted track state indicator is matched. This step is a direct query process. According to the predicted track state indicator selected in the previous step, the pre-calculated abnormal occurrence characteristic value of the track state indicator is retrieved from the established correspondence, and the value is output as the abnormal possibility of the track state indicator at the current time. In order to more specifically illustrate this process, a simplified example can be considered. It is assumed that the track deformation variable indicator of a certain railway line is monitored, the analysis time interval length is set to 10 days, and the moving step length is set to 2 days, so that a time series data sequence and a series of data subsets are obtained. Referring to Table 1, a simple example of judging whether the track deformation variable indicator is abnormal in multiple data subsets and calculating the abnormal occurrence characteristic is shown. The threshold for abnormality judgment is assumed to be that the daily change of the deformation variable is more than 5 mm.

[0047] Table 1: Abnormality statistics table of track state data subsets

[0048] It is assumed that a total of 20 data subsets are analyzed, and the track deformation variable indicator is marked as abnormal in 5 subsets. The total number of track state data subsets is 20. The abnormality occurrence frequency of the track deformation variable indicator in the data subsets is 5. The ratio of the abnormality occurrence frequency to the total number of data subsets is 5 / 20=0.25. The abnormal occurrence characteristic of the track deformation variable indicator is set to 0.25. The fluctuation characteristic (i.e. the average value of the main fluctuation interval) of the track deformation variable indicator is calculated to be 7 days, and the current evaluation time is the 50th day.

[0049] It is assumed that the previous track state data subset (41st day to 50th day) has just been processed at the end of the 50th day. The time difference between the current time (50th day) and the end point of the subset (50th day) is calculated, i.e. the evaluation time difference is 0 days. Then, the deviation amount of the evaluation time difference (0 days) from the fluctuation characteristic (7 days) of the track deformation variable indicator is calculated, i.e. |0-7|=7 days. Since we only consider one track state indicator (deformation variable) as an example, the indicator is the predicted track state indicator corresponding to the minimum deviation amount (which is the only deviation amount here). Finally, the abnormal possibility of the predicted track state indicator (deformation variable) is matched, i.e. the abnormal occurrence characteristic value 0.25 is queried and output. In actual application, multiple track state indicators will be processed simultaneously. Each indicator will have its own independent abnormal occurrence characteristic and fluctuation characteristic. After calculating the deviation amount of the current evaluation time difference from the fluctuation characteristic of each indicator, the indicator with the smallest deviation amount will be selected as the predicted track state indicator.

[0050] The logic of anomaly judgment can be more complex, not limited to a single threshold, and can combine statistical process control methods, such as using mean ± 3 times standard deviation as control limit, or performing anomaly detection based on machine learning models. The calculation of anomaly occurrence features can consider weighted average, and the closer the data subset to the current time, the greater the contribution weight of the anomaly criterion to the feature value, in order to reflect the latest state of the system. The calculation of the evaluation time difference requires high-precision time synchronization services to ensure that all data processing links use a unified clock source. The calculation of the deviation amount can use relative deviation instead of absolute deviation, i.e. (evaluation time difference - fluctuation feature) / fluctuation feature, to eliminate the influence of the magnitude difference of different indicators fluctuation feature. When matching the anomaly possibility, in addition to directly outputting the historical anomaly occurrence feature value, a correction factor based on the deviation amount can also be introduced, for example, the smaller the deviation amount, the higher the matching degree, and the historical feature value is directly used; when the deviation amount is large, the historical feature value is attenuated to a certain extent to reflect the uncertainty of matching. The implementation of the whole process requires a robust data processing pipeline that can handle data delays, window boundary adjustments, etc. For newly commissioned lines or newly deployed monitoring indicators, the lack of historical data, the initial anomaly occurrence feature can be based on the transfer learning of similar lines or engineering experience to set a conservative value, and update quickly with the accumulation of data. While outputting the anomaly possibility, the system usually also records the key basis for the prediction, such as the predicted track state indicator used, its fluctuation feature and deviation amount, so that the operation and maintenance personnel can understand the model decision-making process and perform necessary review.

[0051] Example 5: Based on the predicted track state indicators and their abnormality likelihood, high-risk track sections are screened, which relies on a pre-set risk threshold, for example, track sections with an abnormality likelihood greater than 0.6 are marked as high-risk. The system will automatically generate a high-risk section list containing section numbers, predicted main risk indicators, and their abnormality likelihood estimates. Train operation indicators and environmental disturbance indicators of high-risk track sections are extracted, and the extraction process is performed through spatial and temporal correlation. According to the location information of high-risk sections, the system calls the running indicators of trains passing through the section in the current or near-term plan from the database, such as scheduled speed, load condition, and obtains the current and forecasted environmental disturbance indicators of the section, such as wind speed, precipitation. The train operation safety evaluation value is calculated by combining the operation evaluation model, which takes the indicators extracted in the previous step as input. The input factors are the train operation indicators planned to pass through the high-risk section, and the disturbance factors are the track state indicators of the section (represented by their predicted values and abnormality likelihood) and environmental disturbance indicators. These data are input into the operation evaluation model, which performs calculations and outputs a quantitative train operation safety evaluation value, which reflects the estimated safety level of passing through the high-risk section under the given operation scheme. Based on the train operation safety evaluation value, a railway operation optimization scheme is generated, and the generation logic is reverse, i.e., to ensure that the safety evaluation value is within an acceptable range, the controllable input parameters are adjusted.

[0052] The optimization scheme is specifically embodied in adjusting the train operation scheduling scheme or the track maintenance plan scheme, and adjusting the track maintenance plan scheme focuses on timely repair of infrastructure. The scheme may automatically generate a maintenance work order, suggesting that the originally scheduled track maintenance work in the section next week be performed tonight during the window point, or an additional temporary inspection be added. The scheme will detail the key points of maintenance, such as re-measuring the track geometric dimensions with precision measuring instruments for the predicted abnormal deformation. The process of generating the optimization scheme is not a single mapping, but a multi-objective trade-off decision-making process. The system may generate multiple alternative schemes, such as scheme A suggesting to pass through at a reduced speed, scheme B suggesting to delay the departure by 2 hours to avoid rain, and scheme C suggesting to immediately implement speed limit and schedule track inspection in the early morning of the next day. Each alternative scheme will estimate its effect on the train operation safety evaluation value, and also calculate its impact on operational efficiency, such as increased travel time and decreased throughput. The final decision may be automatically selected by the system according to pre-set priority rules, or multiple alternative schemes and their impact analysis may be pushed to dispatch personnel for final adjudication.

[0053] The technical implementation of adjusting the train operation scheduling scheme needs to interface with the existing train dispatching command system, and the generated adjustment instructions need to conform to the relevant operation specifications and data formats, and can be recognized and executed by the dispatching system. The speed adjustment instruction needs to be specific to the train number, the start and end point coordinates of the limited section and the specific speed limit value. The path adjustment involves route arrangement and signal system linkage, and needs to ensure the feasibility and safety of the path. The adjustment of the track maintenance plan scheme needs to be integrated with the maintenance management information system, and the automatically generated maintenance work order needs to include detailed maintenance elements, such as work site, work item, required materials, recommended work time window and safety precautions. The generation of the scheme has timeliness, and for sudden high-risk early warning, the system needs to have rapid response capability to complete the whole process from risk assessment to scheme generation within a few minutes. For the trend risk, the generated scheme may focus more on the medium and long-term plan adjustment. The effectiveness of the scheme is not fixed, and the system will continuously monitor the state changes of the high-risk section and the effect of the executed optimization scheme, and if the situation worsens or does not meet the expectations, a new round of evaluation and revised scheme will be triggered.

[0054] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0055] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating intelligent operation management of railways, characterized in that, Includes the following steps: Obtain historical railway operation data, extract train operation indicators, track status indicators, and environmental interference indicators, and construct an operation evaluation model based on the train operation indicators, track status indicators, and environmental interference indicators; Set an analysis time interval, perform sliding segmentation on the track status index based on the analysis time interval to obtain a subset of track status data, analyze the subset of track status data, and extract track status fluctuation characteristics and anomaly occurrence characteristics. Obtain the current assessment time, match the corresponding orbital state index according to the orbital state fluctuation characteristics, calculate the probability of anomaly of the orbital state index in combination with the anomaly occurrence characteristics, and output the orbital state index and its probability of anomaly.

2. The railway intelligent operation management evaluation method according to claim 1, characterized in that, The train operation indicators include train speed, train load distribution, and train braking performance; the track condition indicators include track wear degree, track deformation, and track temperature gradient; and the environmental disturbance indicators include wind speed change, humidity change, and temperature change. The method for calculating the degree of track wear includes: Calculate the difference between the actual wear amount and the standard wear amount of the track, and record it as the wear difference value. Calculate the ratio of the wear difference value to the standard wear amount, and set the ratio as the track wear degree. The method for calculating the orbital deformation includes: Acquire track deformation monitoring data, calculate the deformation change per unit time, and set the deformation change as the track deformation variable.

3. The railway intelligent operation management evaluation method according to claim 1, characterized in that, The train operation index is set as the input element, the track status index and environmental interference index are set as interference elements, and the train operation safety assessment value is set as the output element. The method for constructing the operational evaluation model includes: The input and interference elements are converted to time-frequency data. An operation transfer function is constructed based on the converted input and output elements. The interference elements are then introduced, and the operation transfer function is transformed into an operation evaluation model.

4. The railway intelligent operation management evaluation method according to claim 3, characterized in that, Set the first preset duration to the length of the analysis time interval, and the second preset duration to the moving step size; The orbital state indicators are organized into a time-series data sequence. Based on the sliding window technique, starting from the beginning of the time-series data sequence, the window is moved according to the moving step size to extract the orbital state data subset within each window.

5. The railway intelligent operation management evaluation method according to claim 4, characterized in that, For each subset of orbital state data, spectral feature analysis is performed to calculate the power density distribution of orbital state indicators. Power density distribution intervals above a preset threshold are selected, and the average time interval between adjacent intervals is calculated. The average time interval is set as the orbital state fluctuation feature, and a correspondence between orbital state indicators and fluctuation features is established.

6. The railway intelligent operation management evaluation method according to claim 4, characterized in that, The total number of orbital status data subsets is counted, the frequency of anomalies of each orbital status indicator in the data subsets is counted, the ratio of the frequency of anomalies to the total number of data subsets is calculated, the ratio is set as the anomaly occurrence feature, and the correspondence between orbital status indicators and anomaly occurrence features is established.

7. The railway intelligent operation management evaluation method according to claim 1, characterized in that, Obtain the current evaluation time, calculate the time difference between the current time and the end point of the previous orbital state data subset, and record it as the evaluation time difference; Calculate the deviation between the evaluation time difference and the fluctuation characteristics corresponding to each orbital state index, and select the orbital state index corresponding to the smallest deviation as the predicted orbital state index. Based on the correspondence between the orbital state indicators and anomaly occurrence characteristics, the probability of anomalies in the predicted orbital state indicators is matched.

8. The railway intelligent operation management evaluation method according to claim 7, characterized in that, Based on the predicted track condition indicators and their probability of anomalies, high-risk track sections are screened, and train operation indicators and environmental interference indicators of the high-risk track sections are extracted. The train operation safety assessment value is then calculated in conjunction with the operation assessment model.

9. The railway intelligent operation management evaluation method according to claim 8, characterized in that, Based on the train operation safety assessment values, a railway operation optimization plan is generated, and the train operation scheduling plan or track maintenance plan is adjusted.

10. A railway intelligent operation management evaluation system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the railway intelligent operation management evaluation method according to any one of claims 1 to 9.

Images