Comprehensive online monitoring system for electrified railway high-voltage cable

Abstract

The invention discloses a comprehensive online monitoring system for a high-voltage cable of an electrified railway, and particularly relates to the technical field of cable monitoring. The clustering analysis module divides similar groups based on factory and use environment data; the prediction mode selection module evaluates the stability and nonlinearity degree of the environmental data, and selects an optimal prediction method to predict the environmental data; the threshold value judgment module selects a fixed or dynamic threshold value to judge a fault through comparison with similar group data; the dynamic adjustment module optimizes a fault threshold based on historical data and improves the detection precision; the alarm module triggers early warning when the monitoring data exceed a threshold value; according to the method, the operation mode of the cable is accurately matched, the monitoring precision is improved, misjudgment caused by a single data source is avoided, it is ensured that the optimal prediction method is always used under different environmental conditions, the prediction precision of future environmental data is improved, and reliable support is provided for fault judgment.

Description

Technical Field

[0001] The present invention relates to the technical field of cable monitoring, and more particularly to a comprehensive online monitoring system for high-voltage cables in electrified railways. Background Art

[0002] With the rapid development of electrified railway transportation, the stability, reliability, and safety of its power supply system are crucial to railway operations. High-voltage cables, as a core component of electrified railway power supply systems, are susceptible to external damage, insulation aging, overvoltage shocks, and misoperation during long-term operation. These failures can lead to faults such as short circuits, disconnections, partial discharges, and multiple grounding points. These faults can not only affect the stability of railway power supply but also damage power supply equipment and even disrupt the normal operation of trains. Therefore, real-time monitoring of the operating status of high-voltage cables and early detection of potential hazards are crucial to improving the operational safety and maintenance efficiency of power systems.

[0003] At present, high-voltage cable online monitoring technology has been widely used in the power industry to obtain the real-time operating status of the equipment and realize early warning through data analysis. Existing online monitoring systems often rely on fixed thresholds when judging faults, which makes it difficult to adapt to environmental changes: Existing online monitoring systems usually make fault judgments based on fixed thresholds, that is, when a certain monitoring indicator (such as partial discharge amplitude, temperature, current) exceeds the set value, the system triggers an alarm. However, the environment along the railway is complex, and various environmental factors (such as temperature, humidity, current load) will affect the operating status of the high-voltage cable, making it difficult for a single fixed threshold to be applied to different scenarios, which may lead to false alarms or missed reports. Therefore, the present invention proposes a comprehensive online monitoring system for high-voltage cables on electrified railways, in order to improve the operating safety and maintenance efficiency of the power system. Summary of the Invention

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] An integrated online monitoring system for high-voltage cables in electrified railways, comprising a data acquisition module, a cluster analysis module, a prediction method selection module, a threshold determination module, a dynamic adjustment module, and an alarm module;

[0006] The data acquisition module is used to obtain the environmental data and operating status data of the high-voltage cable;

[0007] The cluster analysis module is used to perform cluster analysis based on the factory and usage environment data of high-voltage cables, and divide the cables into similar usage groups;

[0008] The prediction method selection module is used to evaluate the stability and nonlinearity of environmental data, select the preset corresponding applicable prediction method based on the evaluation results, and predict the operating data group at the next moment;

[0009] The threshold judgment module is used to predict the results and cable similarity groups, perform judgment analysis, and select fixed threshold or dynamic threshold method for fault judgment;

[0010] The dynamic adjustment module is used to dynamically adjust the fault judgment threshold based on historical data to optimize the fault detection accuracy;

[0011] The alarm module is used to trigger an early warning when the monitoring data exceeds a fixed threshold or a dynamic threshold.

[0012] In a preferred embodiment, before cluster analysis, the high-voltage cable factory data is converted into factory vectors, the usage environment data is converted into environment vectors, and the factory vectors and environment vectors are aggregated into online monitoring vectors.

[0013] In a preferred embodiment, the cluster analysis uses a K-Means clustering algorithm or a DBSCAN clustering algorithm to cluster the current high-voltage cable with the remaining high-voltage cables and divide the cables into similar groups.

[0014] In a preferred embodiment, when evaluating the stability and nonlinearity of environmental data, a stability index for measuring the stability of environmental data and a nonlinearity index for measuring the nonlinearity of the high-voltage cable are generated respectively.

[0015] In a preferred embodiment, selecting a preset corresponding applicable prediction method according to the evaluation result and predicting the operation data set at the next moment refers to:

[0016] A pre-trained machine learning model, namely the convolutional neural network model, is used. The stability index and the nonlinear index are used as input data, and the type of prediction method is used as output data. The prediction methods include time series prediction method and regression prediction method.

[0017] In a preferred embodiment, the logic for obtaining the stability index is:

[0018] Calculate the standard deviation of one type of environmental data within a preset time window:

[0019] X i is the data value corresponding to the sampling point i of one type of environmental data in the preset time window, μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, and σ is the standard deviation of one type of environmental data in the preset time window;

[0020] Calculate the coefficient of variation of the data to normalize the degree of data fluctuation:

[0021] CV is the coefficient of variation of the data;

[0022] The calculation formula of the stability coefficient is:

[0023] SI=e -α.CV ; α is the preset non-zero adjustment coefficient, SI is the stability coefficient, and among all categories of environmental data, the maximum value of the stability coefficient is taken as the stability index.

[0024] In a preferred embodiment, the logic for obtaining the nonlinear index is:

[0025] Get the first-order difference sequence of one type of environmental data within the preset time window, that is, the rate of change:

[0026] ΔX i =X i+1 -X i ;X i is the data value corresponding to the sampling point i of one type of environmental data within the preset time window, X i+1 is the data value corresponding to the sampling point i+1 of one type of environmental data within the preset time window, ΔX i The change rate of sampling point i+1 of one type of environmental data relative to sampling point i within the preset time window;

[0027] Calculate the coefficient of variation of the difference:

[0028] σ ΔX and μ ΔX are the standard deviation and mean of the first-order difference sequence, CV Δ is the coefficient of variation of the difference;

[0029] Calculate the autocorrelation coefficient:

[0030] μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, k is the preset time lag order, X i+k is the data value corresponding to the sampling point i+k of one type of environmental data within the preset time window, and ACF(k) is the autocorrelation coefficient of one type of environmental data within the preset time window;

[0031] The nonlinear coefficient calculation formula is:

[0032] NLI=β1·CV △ +β2·(1-|ACF(1)|); β1 and β2 are both preset non-zero proportional coefficients, and their sum is one. ACF(1) is the autocorrelation coefficient corresponding to the value of k equal to 1. NLI is the nonlinear coefficient. Among all categories of environmental data, the maximum value of the nonlinear coefficient is taken as the nonlinear index.

[0033] In a preferred embodiment, the threshold determination module is used to predict the results and cable similarity groups, perform determination analysis, and select a fixed threshold or a dynamic threshold for fault determination.

[0034] The current environmental data group composed of the prediction results of multiple types of environmental data is compared with the group environmental data group of similar cable groups under current environmental conditions, and the Euclidean distance between the current environmental data group and the group environmental data group is calculated. If the Euclidean distance is greater than the preset deviation threshold, the dynamic threshold method is used for fault judgment. If the Euclidean distance is less than or equal to the preset deviation threshold, the fixed threshold method is used for fault judgment.

[0035] In a preferred embodiment, the dynamic adjustment module is used to dynamically adjust the fault determination threshold based on historical data, which includes:

[0036] Obtain the mean value of the operating status data of category h under the same historical environmental conditions as the current environmental conditions for similar groups. h , the standard deviation is σ h , and the preset optimization coefficient k h , and then substitute into the following formula:

[0037] T dynamic =μ h +k h ·σ h ;T dynamic Indicates the dynamic threshold of the running status corresponding to category h.

[0038] Technical effects and advantages of the present invention:

[0039] The present invention not only collects the operating status data of the cable (such as current, voltage, partial discharge signal, etc.), but also combines environmental data (such as temperature, humidity, electromagnetic interference, etc.) to form a comprehensive data set to fully reflect the operating status of the high-voltage cable. By clustering and analyzing the factory data, historical usage data, and real-time environmental data, the operating mode of the cable is accurately matched, the monitoring accuracy is improved, and the misjudgment caused by a single data source is avoided. The present invention introduces K-Means or DBSCAN cluster analysis, and divides cables with similar operating characteristics into similar groups based on the factory data and usage environment data of the high-voltage cable. When performing status assessment, the current operating status of the cable is compared with the historical data of similar groups to ensure the rationality of threshold setting and predictive analysis. Compared with traditional methods, it can provide personalized status assessment and fault prediction for cables of different models and different usage environments, thereby improving the adaptability and accuracy of monitoring.

[0040] This invention proposes a stability index (to measure data volatility) and a nonlinearity index (to measure data complexity) to evaluate environmental data characteristics from different dimensions. Using a convolutional neural network model, it automatically selects either a time series prediction method or a regression prediction method: For stable data with high linearity, time series prediction is used to improve prediction accuracy. For unstable data with high nonlinearity, regression prediction is used to improve adaptability in complex environments. This method ensures that the optimal prediction method is always used under different environmental conditions, improving the prediction accuracy of future environmental data and providing reliable support for fault diagnosis.

[0041] In this method, if the current environmental data deviates slightly from the environmental data of a similar historical group, indicating a stable environment, a fixed threshold can be used directly for fault diagnosis, reducing the computational burden and improving detection efficiency. If the current environmental data deviates significantly from the environmental data of a similar historical group, indicating significant environmental changes, a dynamic threshold should be used for fault diagnosis to adapt to environmental changes and improve detection accuracy. This method can dynamically adapt to different environmental conditions and significantly reduces false positives and missed negatives compared to traditional fixed threshold methods, thereby improving fault detection accuracy.

[0042] The present invention dynamically adjusts the fault determination threshold by calculating the mean and standard deviation of historical data of similar groups under the same environmental conditions. Through precise monitoring and intelligent prediction, it can avoid premature cable replacement due to unknown faults, extend the service life of cables, reduce replacement costs, improve power supply reliability, and reduce sudden power accidents. Through intelligent monitoring and remote alarms, early warnings can be issued before faults occur, reducing sudden power outages caused by cable faults and improving the stability of the power supply system. Traditional cable inspections rely on manual labor. The present invention provides online monitoring + intelligent analysis, which can reduce the frequency of manual inspections, improve operation and maintenance efficiency, and reduce labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0044] Figure 1 The schematic diagram of the integrated online monitoring system for high-voltage cables in electrified railways according to the present invention is shown. DETAILED DESCRIPTION

[0045] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] Reference Figure 1The following examples were obtained:

[0047] Example 1:

[0048] With the rapid development of electrified railway transportation, the stability, reliability, and safety of its power supply system are crucial to railway operations. High-voltage cables, as a core component of electrified railway power supply systems, are susceptible to external damage, insulation aging, overvoltage shocks, harmonics, and other factors during long-term operation. These factors can lead to faults such as short circuits, overheating, partial discharge, and multiple grounding points. These faults can not only affect the stability of railway power supply but also damage power supply equipment and even disrupt the normal operation of trains. Therefore, real-time monitoring of the operating status of high-voltage cables and early detection of potential hazards are crucial to improving the operational safety and maintenance efficiency of power systems.

[0049] Currently, high-voltage cable online monitoring technology has been widely used in the power industry to obtain the real-time operating status of equipment and provide early warning through data analysis. However, the existing online monitoring system still has the following problems:

[0050] Fault detection relies on fixed thresholds, making it difficult to adapt to environmental changes: Existing online monitoring systems typically use fixed thresholds for fault detection. This means that when a monitoring indicator (such as partial discharge amplitude, temperature, or current) exceeds a set value, the system triggers an alarm. However, the complex environment along railways, with multiple environmental factors (such as temperature, humidity, and current load) affecting the operating status of high-voltage cables, makes a single fixed threshold difficult to apply to different scenarios, potentially leading to false alarms or missed alarms.

[0051] Lack of intelligent prediction capabilities based on environmental conditions: Most existing online monitoring systems rely on historical experience or single statistical models for fault diagnosis, lacking the ability to intelligently predict real-time environmental conditions. The operating status of high-voltage cables is affected by multiple factors, and existing technologies fail to intelligently select appropriate prediction methods based on the stability and nonlinear characteristics of environmental data, resulting in low prediction accuracy.

[0052] Failure to conduct differentiated monitoring based on the characteristics of different high-voltage cable groups: Due to differences in operating environments, factory processes, and other factors, high-voltage cables vary in reliability and failure trends. Existing online monitoring systems often employ a unified monitoring strategy, failing to conduct personalized analysis and threshold adjustments for different equipment groups, resulting in insufficient monitoring accuracy.

[0053] In summary, existing high-voltage cable online monitoring technology still has significant room for improvement in terms of threshold determination, prediction method selection, and adaptability to different environments and equipment characteristics. Therefore, a high-voltage cable online monitoring system is urgently needed that can intelligently predict faults based on environmental data characteristics, optimize fault determination based on similarity group analysis, and dynamically adjust thresholds. This can improve the safety, fault prediction capabilities, and operation and maintenance efficiency of electrified railway power supply systems.

[0054] The purpose of this invention is to provide a comprehensive online monitoring system for high-voltage cables in electrified railways to address existing issues in high-voltage cable fault detection, such as the limitations of fixed thresholds, insufficient predictive capabilities, and inability to adapt to the characteristics of diverse equipment groups. Its core objectives include:

[0055] Construct a comprehensive online monitoring device for high-voltage cables to realize multi-parameter data collection: Design a comprehensive online monitoring device for high-voltage cables, including a high-frequency pulse current sensor, sheath protector sensor, ground loop current transformer, temperature sensor and data communication module, to realize real-time online monitoring of parameters such as partial discharge, ground loop current, and temperature, providing data support for equipment status assessment.

[0056] Use cluster analysis technology to divide high-voltage cables into similar groups: Use K-Means or DBSCAN clustering algorithms to analyze the factory data and usage environment data of high-voltage cables, construct online monitoring vectors, and divide high-voltage cables with similar operating status into the same group to optimize the monitoring strategy.

[0057] Based on the stability and nonlinearity of environmental data, the system intelligently selects a prediction method: It calculates the stability index and nonlinearity index to assess the volatility and complexity of environmental data. Using a pre-trained convolutional neural network (CNN) model, the system automatically selects the most appropriate prediction method (such as time series prediction or regression prediction) based on the stability and nonlinearity indexes of environmental data, and predicts the next set of operating data. Based on the characteristics of similar groups and the prediction results, it selects a fixed or dynamic threshold for fault diagnosis.

[0058] Calculate the Euclidean distance between the current environmental data set and similar groups of environmental data sets: If the distance is small (i.e., the environment is stable), a fixed threshold is used for fault diagnosis to reduce computational complexity. If the distance is large (i.e., the environment is changing significantly), a dynamic threshold is used for fault diagnosis to improve adaptability. The fault diagnosis threshold is dynamically adjusted based on historical data to improve detection accuracy.

[0059] When monitoring data exceeds a set threshold, the alarm module triggers a warning signal and provides a trend analysis report, prompting operations personnel to take appropriate measures. Combining fiber optic, RS485, LoRa / 5G, and other communication technologies enables remote data transmission and real-time monitoring, improving fault response speed.

[0060] The high-voltage cable integrated online monitoring device of the present invention is described by way of example, including:

[0061] High-frequency pulse current sensor: used to couple partial discharge signals and monitor cable insulation status;

[0062] Sheath protector sensor: used to detect leakage current and evaluate the health of the cable sheath;

[0063] Ground loop current transformer: used to monitor ground current and determine whether there is a ground fault;

[0064] Temperature sensor: used to monitor cable temperature in real time and assess temperature anomalies;

[0065] Data communication module: supports optical fiber communication and LoRa / 5G wireless communication to achieve remote data transmission.

[0066] The invention discloses a comprehensive online monitoring system for high-voltage cables in electrified railways, comprising a data acquisition module, a cluster analysis module, a prediction mode selection module, a threshold determination module, a dynamic adjustment module, and an alarm module; the modules are communicatively connected with each other.

[0067] The data acquisition module is used to acquire environmental and operational data from high-voltage cables, providing foundational data for subsequent data analysis and fault diagnosis. Collected data includes, but is not limited to, environmental data such as temperature, humidity, air pressure, and electromagnetic interference intensity. Operational data such as current, voltage, ground loop current, partial discharge signals, sheath leakage current, and cable temperature. This ensures the system can capture real-time operating information about high-voltage cables, providing a reliable data source for analyzing their health. Through long-term, continuous monitoring, it is possible to identify abnormal cable trends, providing essential data for predictive analysis and early warning.

[0068] The cluster analysis module is used to perform cluster analysis based on the factory and operating environment data of high-voltage cables, dividing cables into groups with similar usage to provide a reference standard for subsequent predictions and threshold determination. Key steps: Data vectorization: Factory data is converted into factory vectors (including the cable's insulation grade and manufacturing parameters). Operating environment data is converted into environmental vectors (including operating environment temperature, humidity, voltage, current, etc.). Both are combined into online monitoring vectors to form a unified calculation standard. Cluster analysis: Using the K-Means clustering algorithm or the DBSCAN clustering algorithm, devices with similar status are divided into similar groups based on the environmental and operating parameters of different high-voltage cables. The characteristic means and historical operating patterns of similar groups are determined to provide a benchmark for subsequent predictions and threshold determination. Differentiating the operating status of different high-voltage cables avoids applying the same monitoring standards to devices in different environments, improving the accuracy of fault determination. The historical operating data of similar groups provides a more reasonable reference for prediction analysis and improves the accuracy of early warnings.

[0069] The prediction method selection module evaluates the stability and nonlinearity of environmental data. Based on the evaluation results, it selects a pre-set, applicable prediction method and predicts the next set of operating data. The main steps are: Calculating the stability index: This module measures the data's volatility by calculating the standard deviation and coefficient of variation of the environmental data. Calculating the nonlinearity index: This module uses methods such as first-order differences and autocorrelation analysis to assess the degree of nonlinearity in the data and determine whether complex trends exist. Selecting a prediction method: If the data is stable and the trend is clear, a time series prediction method (such as ARIMA or LSTM) is used. If the data is unstable and the changes are complex, a regression prediction method (such as multivariate regression or XGBoost) is used. Predicting the next set of operating data provides input data for subsequent threshold determination. Dynamic analysis of environmental data allows the selection of an appropriate prediction method, improving data prediction accuracy, and predicting future environmental changes. This helps to identify potential cable faults in advance and enhances fault warning capabilities.

[0070] The threshold judgment module is used to analyze prediction results and cable similarity groups, and select either a fixed threshold or a dynamic threshold for fault diagnosis. The main steps are: Calculate the Euclidean distance between the current environmental data and the environmental data of the similar group. If the distance is small (i.e., the environmental change is small), a fixed threshold is used for fault diagnosis. If the distance is large (i.e., the environmental change is significant), a dynamic threshold is used for fault diagnosis. Fault diagnosis: If the data exceeds the set threshold, it is determined to be an abnormality and is handled by the alarm module. Selecting the appropriate threshold judgment method based on environmental changes improves fault detection accuracy. This avoids the inadaptability of fixed thresholds and enhances the system's adaptability to different environments.

[0071] The dynamic adjustment module dynamically adjusts the fault determination threshold based on historical data to optimize fault detection accuracy. This dynamic adjustment improves the adaptability of the threshold, reduces false positives and false negatives, and improves system reliability. The alarm module triggers early warnings when monitoring data exceeds fixed or dynamic thresholds, providing timely alarms, improving response speed, and preventing high-voltage cable accidents.

[0072] Before cluster analysis, the high-voltage cable's factory data is converted into factory vectors, and the operating environment data is converted into environmental vectors. The factory and environmental vectors are then aggregated into online monitoring vectors. The core purpose of this step is to establish a unified data format, enabling consistent processing of data from different sources and providing a mathematical foundation for subsequent cluster analysis. The factory vector is a vectorized representation of the cable's initial characteristic data at the factory, primarily including the cable's manufacturing parameters and quality indicators. The data source is factory inspection data provided by the manufacturer, including material properties and rated operating parameters. The environmental vector represents the operating environmental conditions of the high-voltage cable, including factors such as temperature, humidity, and electromagnetic interference. The data source is real-time environmental data acquired by the data acquisition module. This vector reflects the dynamic changes in the cable's environment and can be used to analyze the impact of environmental factors on the cable's operating status. Combined with the factory vector, it can be used to evaluate cable performance changes under different environmental conditions. The online monitoring vector is a comprehensive vector that represents the current status of the high-voltage cable by combining the factory and environmental vectors. The data source is factory data, operating data, and monitoring data. Structuring the status data of different high-voltage cables into a unified vector facilitates mathematical calculations, similarity analysis, and clustering. Combining design parameters and operating environment enables more accurate fault diagnosis. Relying solely on factory parameters cannot reflect the cable's true condition after long-term operation; relying solely on environmental data cannot determine whether cable anomalies are caused by design flaws. Combining factory and environmental data vectors allows for a more comprehensive assessment of the cable's current health. Based on online monitoring vectors, K-Means and DBSCAN clustering algorithms can be used to group devices with similar operating conditions. The operating data of each cable type can be compared with similar groups, improving the accuracy of fault diagnosis.

[0073] Cluster analysis uses the K-Means clustering algorithm or the DBSCAN clustering algorithm to cluster the current high-voltage cables with the rest of the high-voltage cables and divide the cables into similar groups.

[0074] The K-Means clustering algorithm is an unsupervised learning method based on partitioning. Its goal is to divide data points into K different clusters so that the data points in each cluster are as similar as possible to each other, while the data points between different clusters are as different as possible.

[0075] Steps of the K-Means algorithm: I. Initialization: Set the number of clusters K (i.e., the number of similar groups of high-voltage cables to be divided). Randomly select K points in the dataset as the initial cluster centers. II. Calculate the distance from each data point to the cluster center: Calculate the Euclidean distance between the data point and each cluster center. III. Assign data points: Assign each high-voltage cable to the cluster where the nearest cluster center is located. IV. Update the cluster center: Calculate the mean of all data points in each cluster and use the mean as the new cluster center. V. Repeat the iteration: Repeat steps II to IV until: the cluster centers no longer change significantly, i.e., converge, or reach the set maximum number of iterations. VI. Final output: Generate K different groups of high-voltage cables. The operating characteristics of the devices in each group are similar and can be used for subsequent prediction analysis and threshold setting. The value of K can be determined by the elbow method. The K-Means algorithm is applicable to the case where the state data distribution of high-voltage cables is relatively uniform.

[0076] DBSCAN is a density-based clustering method that can discover clusters of any shape and can identify noise points. Compared with K-Means, it does not require specifying the number of clusters KK and is applicable to the case where the state distribution of high-voltage cables is uneven or contains outliers.

[0077] Steps of the DBSCAN algorithm: (1) Parameter setting: Neighborhood radius ε: That is, the maximum distance within which data points are considered density-connected. Minimum number of samples MinPts: That is, there need to be at least MinPts points around a data point to form a cluster. (2) Traverse data points: For each high-voltage cable data point: Calculate the number of data points within its ε-neighborhood. If the number of points in the neighborhood ≥ MinPts, then this point is a core point and forms a new cluster. If the number of points in the neighborhood < MinPts and ≥ 1, then this point is a boundary point, belonging to a cluster but not able to form a cluster independently. If the number of points in the neighborhood = 0, then this point is a noise point and will not be assigned to any cluster. (3) Expand the cluster: Taking the core point as the center, find all directly density-reachable points and continue to expand outward until the entire cluster is formed. (4) Mark noise points: For the points that are not assigned to any cluster, mark them as noise points. (5) Repeat the above steps: Traverse all data points until all points are classified or marked as noise points. (6) Final output: Generate high-voltage cable groups of any shape. The operating characteristics of the devices in each group are similar, and the noise points are treated as abnormal data. The DBSCAN algorithm is applicable to the case where the device data distribution is uneven and there are abnormal data. It does not require specifying the number of clusters KK and has stronger adaptability to different operating states of high-voltage cables. It is suitable for discovering device groups with high density and can identify devices in abnormal states at the same time, facilitating subsequent analysis.

[0078] When evaluating based on the stability and nonlinearity of environmental data, a stability index for measuring the stability of the environmental data and a nonlinearity index for measuring the nonlinearity of the high-voltage cable are generated respectively.

[0079] Selecting a preset corresponding applicable prediction method based on the evaluation results and predicting the operating data group at the next moment means: using a pre-trained machine learning model, namely a convolutional neural network model, taking the stability index and the nonlinear index as input data, and taking the type of prediction method as output data. There are two types of prediction methods: time series prediction method and regression prediction method.

[0080] In this invention, to select the most appropriate prediction method and predict the next set of operating data, the system uses a convolutional neural network model. It takes the stability index and nonlinearity index of environmental data as input and the type of prediction method as output, automatically determining the applicable prediction method. These prediction methods primarily include time series prediction and regression prediction. The core principles are as follows:

[0081] Characteristics of input data: When predicting future operating data, it is necessary to first evaluate the characteristics of the current environmental data. This evaluation is based on two key indicators: Stability Index: This is used to measure whether the changes in environmental data are stable. If the environmental data has fluctuated slightly over a period of time, it is considered to have high stability. Nonlinearity Index: This is used to measure whether the environmental data has complex nonlinear characteristics. If the data change pattern is relatively regular, the degree of nonlinearity is low; if the data changes erratically or is difficult to describe with a simple mathematical model, the degree of nonlinearity is high. These indices are calculated based on the historical changes in environmental data and are obtained through a series of statistical analysis methods. After the calculation is completed, the system inputs the stability index and nonlinearity index as feature data into the convolutional neural network model.

[0082] Training of the machine learning model: When selecting the prediction method, the system uses a convolutional neural network model. The training process of this model is as follows:

[0083] Data preparation: Collect a large amount of historical environmental data, calculate the stability index and nonlinear index within the corresponding time window, and mark the most suitable prediction method at that time.

[0084] Feature extraction: Use convolutional neural networks to perform feature analysis on input data and extract important information from data patterns to more accurately distinguish different environmental features.

[0085] Model training: Based on a large amount of labeled data, the convolutional neural network is trained to enable it to learn the mapping relationship between the stability index and nonlinear index and the optimal prediction method.

[0086] Model validation and optimization: Use the test data set to verify the accuracy of the model and continuously optimize it based on the error situation to improve the accuracy of the prediction method selection.

[0087] Through training, the neural network model can automatically determine which prediction method is most suitable for the current environmental data based on the input stability index and nonlinear index.

[0088] Prediction Method Selection: When the system needs to predict the next set of operating data, it first calculates the stability index and nonlinearity index of the current environmental data. These indices are then fed into a trained neural network model. Based on the patterns learned from past data, the model automatically outputs an appropriate prediction method. Specifically, when the stability index is high and the nonlinearity index is low, the environmental data's changing pattern is relatively stable, and the data trend can be described by a regular mathematical model. This is a suitable method for time series prediction, which can predict future operating data based on past data trends. When the stability index is low and the nonlinearity index is high, the environmental data's changing pattern is complex, potentially influenced by multiple factors and exhibiting nonlinear characteristics. This is a suitable method for regression prediction, which leverages the correlations between multiple environmental factors to construct a nonlinear model for more accurate predictions of future operating data. After the model outputs the prediction method, the system calls the corresponding prediction model and uses it to calculate the next set of operating data, ultimately generating the prediction result.

[0089] Application of Forecasting Methods: This paper uses both time series and regression forecasting methods, each of which has different application scenarios. The time series forecasting method is suitable for scenarios where data changes are relatively stable and can be used to make predictions based on historical trends. The regression forecasting method is suitable for scenarios where data changes are complex and can establish mathematical relationships based on multiple environmental factors, thereby improving forecast accuracy.

[0090] When evaluating based on the stability and nonlinearity of environmental data, a stability index for measuring the stability of the environmental data and a nonlinearity index for measuring the nonlinearity of the high-voltage cable are generated respectively.

[0091] The logic for obtaining the stability index is:

[0092] Calculate the standard deviation of one type of environmental data within a preset time window:

[0093] X i is the data value corresponding to the sampling point i of one type of environmental data in the preset time window, μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, and σ is the standard deviation of one type of environmental data in the preset time window;

[0094] Calculate the coefficient of variation of the data to normalize the degree of data fluctuation:

[0095] CV is the coefficient of variation of the data;

[0096] The calculation formula of the stability coefficient is:

[0097] SI=e -α·CV ; α is the preset non-zero adjustment coefficient, SI is the stability coefficient, and among all categories of environmental data, the maximum value of the stability coefficient is taken as the stability index.

[0098] When monitoring the operating status of high-voltage cables, the stability of environmental data directly impacts the accuracy of fault prediction. Minimal fluctuations in environmental data indicate a stable environment and relatively predictable cable operation. Conversely, large fluctuations indicate an unstable environment and significant influence of external factors. Therefore, a mathematical formula is needed to measure the degree of environmental data fluctuation and convert it into a stability index for subsequent fault diagnosis.

[0099] The calculation formula of the stability index consists of three core parts:

[0100] Calculate the standard deviation of environmental data: This measures the fluctuation of data within a time window. The standard deviation indicates the magnitude of data fluctuation. A large value indicates drastic data fluctuations and an unstable environment; a small value indicates minimal data fluctuations and a stable environment. The standard deviation can be used to quantify the degree of data dispersion and assess the dynamic nature of environmental changes.

[0101] Calculating the coefficient of variation (CV) of data: This normalizes data fluctuations to ensure that data of different dimensions can be compared. The CV is a normalized metric for fluctuations that eliminates the influence of data of different dimensions, allowing direct comparison of data from different environmental variables. Using only the standard deviation (SD) can lead to incomparable results due to varying ranges of environmental data. For example, the standard deviations of temperature and current can vary significantly, but using the CV standard deviations normalizes them, ensuring data comparability.

[0102] Calculate the stability coefficient: The data is converted using an exponential decay function so that the stability coefficient is within the range of (0, 1] for ease of use. The maximum value of the stability coefficient is taken as the stability index. The exponential decay function is used to convert the coefficient of variation into a stability index in the range of (0, 1]. A larger coefficient of variation indicates more drastic data fluctuations and a more unstable environment. In this case, the stability index approaches 0. A smaller coefficient of variation indicates less data fluctuations and a more stable environment. In this case, the stability index approaches 1. The role of exponential decay is to enhance discrimination, ensuring that when data fluctuations are small, the index changes smoothly, and when data fluctuations are large, the index changes faster, thereby increasing sensitivity.

[0103] In environmental data analysis, multiple categories of environmental data are typically monitored, such as temperature, humidity, electromagnetic interference, ground resistance, and sheath leakage current. Different categories of data may have different fluctuation characteristics, so a representative value must be selected as the final stability index. The maximum stability coefficient across all categories of environmental data is selected based on the following considerations:

[0104] Maintaining the strictest stability standards: If one of the multiple environmental factors fluctuates significantly, the overall environment is unstable. Maximizing the value ensures that the system doesn't misjudge the entire environment as stable due to low volatility in some data.

[0105] The logic behind cable operation is consistent with the following: The operating status of high-voltage cables is affected by multiple factors, and the most unstable factor often determines the risk of cable failure. For example, if temperature fluctuations are small but electromagnetic interference is strong, the cable may still experience anomalies. Therefore, the most unstable factor should be used as the final evaluation criterion.

[0106] Adapting to weightings of different environmental factors: In some cases, different types of environmental data may have different weights on the impact of cables. For example, temperature changes may have a greater impact on some cables, while humidity changes may have a greater impact on others. Taking the maximum value ensures that the most critical influencing factors are selected, enhancing adaptability to different operating conditions.

[0107] The logic for obtaining the nonlinear index is:

[0108] Get the first-order difference sequence of one type of environmental data within the preset time window, that is, the rate of change:

[0109] ΔX i =X i+1 -X i ;X i is the data value corresponding to the sampling point i of one type of environmental data within the preset time window, X i+1 is the data value corresponding to the sampling point i+1 of one type of environmental data within the preset time window, ΔX iThe change rate of sampling point i+1 of one type of environmental data relative to sampling point i within the preset time window;

[0110] Calculate the coefficient of variation of the difference:

[0111] σ ΔX and μ ΔX are the standard deviation and mean of the first-order difference sequence, CV Δ is the coefficient of variation of the difference;

[0112] Calculate the autocorrelation coefficient:

[0113] μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, k is the preset time lag order, X i+k is the data value corresponding to the sampling point i+k of one type of environmental data within the preset time window, and ACF(k) is the autocorrelation coefficient of one type of environmental data within the preset time window;

[0114] The nonlinear coefficient calculation formula is:

[0115] NLI=β1·CV Δ +β2·(1-|ACF(1)|); β1 and β2 are both preset non-zero proportional coefficients, and their sum is one. ACF(1) is the autocorrelation coefficient corresponding to the value of k equal to 1. NLI is the nonlinear coefficient. Among all categories of environmental data, the maximum value of the nonlinear coefficient is taken as the nonlinear index.

[0116] The nonlinearity of environmental data reflects the complexity of data changes over time. To quantify nonlinear characteristics, this paper uses two core indicators:

[0117] Coefficient of Variation of Differences: Measures the instability of the data's rate of change. First-order differences are used to calculate the rate of change between two consecutive data points, describing the local fluctuations in the data. If the data is linear, the first-order differences will change relatively smoothly; if the data is nonlinear, the first-order differences will show irregular fluctuations. The coefficient of variation measures the degree of fluctuation in the first-order differences, that is, the instability of the data's rate of change. A small coefficient of variation indicates a relatively stable rate of change and a relatively linear trend; a large coefficient of variation indicates a highly fluctuating rate of change and a relatively nonlinear trend.

[0118] Autocorrelation coefficient: This measures the correlation between current data and past data, reflecting the data's predictability. These two metrics, respectively, measure the data's local variation and overall trend characteristics, and together form the basis for calculating the nonlinearity index. The autocorrelation coefficient reflects the degree of similarity between current and past data. A large autocorrelation coefficient indicates strong predictability and a more linear trend; a small autocorrelation coefficient indicates less predictability and a more nonlinear trend.

[0119] The coefficient of variation and the autocorrelation coefficient are combined through a weighted summation to determine the degree of nonlinearity in the data. A larger coefficient of variation indicates a more volatile rate of change in the data and a higher degree of nonlinearity; a smaller autocorrelation coefficient indicates a lack of regularity in the data and a higher degree of nonlinearity. By weighted summation, the influence of these two indicators is relatively balanced, allowing the nonlinearity index to comprehensively measure the degree of nonlinearity in the data.

[0120] When monitoring environmental data, multiple environmental factors are typically analyzed, such as temperature, humidity, electromagnetic interference, sheath leakage current, and ground resistance. Different types of data may exhibit different nonlinear characteristics, so it's important to select an indicator that best represents the complexity of the current environment. The maximum nonlinear coefficient across all environmental data types is selected based on the following considerations:

[0121] Avoid misleading low nonlinearity data: If the nonlinearity index of most environmental variables is low, but one variable is high, it indicates that this variable may have the greatest impact on the system's operating state. Taking the maximum value ensures that the system does not underestimate the overall nonlinearity of the environment due to the influence of a single linear variable.

[0122] Meets the practical needs of high-voltage cable fault prediction: The operating status of high-voltage cables is affected by a combination of multiple environmental factors, but the most unstable of these factors often determines the system's failure risk. For example, if the nonlinearity index for temperature and humidity is low, but the nonlinearity index for electromagnetic interference is high, electromagnetic interference may be the primary influencing factor and warrants special attention. Maximizing the value ensures that the environmental factor with the most nonlinear characteristics is identified, improving the accuracy of fault prediction.

[0123] Enhance the adaptability of the prediction model: The choice of prediction method depends on the nonlinear characteristics of the environment. If only the mean or median is used, the nonlinear characteristics of some environmental variables may be weakened. Taking the maximum value ensures that the prediction method is applicable to the most complex situations, allowing the system to select the optimal prediction strategy in any environment.

[0124] For example, the stability index reflects the most stable or most unstable variable in the environmental data, while the nonlinearity index reflects the most predictable or least predictable variable in the environmental data. Because different environmental parameters affect the system differently, some changes are more significant than others, so it makes sense to take the maximum value. For example, if temperature has the largest variation, meaning it has the greatest impact on cable stability, the stability index would be based on temperature. If humidity has the strongest nonlinearity, meaning its variation pattern is the most unpredictable, the nonlinearity index would be based on humidity. By taking the maximum value, the system focuses on the most significant influencing factors and is not distracted by smaller environmental fluctuations.

[0125] The operating environment parameters of high-voltage cables are not completely independent; rather, they exhibit certain correlations. For example, changes in temperature can lead to changes in humidity, which are physically related. Cable discharge can be affected by a combination of humidity, temperature, and electromagnetic interference. Therefore, even though the stability index and nonlinearity index are derived from different environmental variables, they still represent the overall environmental characteristics, thus influencing the selection of prediction methods for the entire system.

[0126] The threshold judgment module is used to predict the results and cable similarity groups, perform judgment analysis, and select fixed threshold or dynamic threshold method for fault judgment.

[0127] The current environmental data group composed of the prediction results of multiple types of environmental data is compared with the group environmental data group of similar cable groups under current environmental conditions, and the Euclidean distance between the current environmental data group and the group environmental data group is calculated. If the Euclidean distance is greater than the preset deviation threshold, the dynamic threshold method is used for fault judgment. If the Euclidean distance is less than or equal to the preset deviation threshold, the fixed threshold method is used for fault judgment.

[0128] Traditional high-voltage cable monitoring systems typically use fixed thresholds for fault diagnosis. For example, if the partial discharge signal exceeds a certain value, an alarm is triggered. If the temperature exceeds a set upper limit, an alarm is triggered. However, fixed thresholds have limitations: they may not be applicable under different environmental conditions. High-voltage cables may operate differently in different environments (such as humidity and current load), and fixed thresholds cannot adapt to dynamic environmental changes, easily leading to false alarms or missed alarms.

[0129] The same type of cable may have different fault thresholds in different usage scenarios (such as mountainous areas and high-temperature regions). Therefore, relying solely on fixed thresholds will result in poor generalization of the detection system. The present invention uses similarity group analysis to calculate the Euclidean distance between current environmental data and historical similar group data to determine whether the current environment is in a known operating mode, thereby determining whether to use a fixed threshold or a dynamic threshold for fault determination.

[0130] If the current environmental data is very close to the environmental data of a similar group in the past, it means that the current operating status of the cable is consistent with the historical operating conditions, and a fixed threshold can be used directly for fault diagnosis. This can reduce computing costs and improve the stability of fault diagnosis.

[0131] If there is a significant deviation between the current environmental data and the environmental data of similar groups in the past, it means that the current cable operating environment has changed. Using dynamic thresholds means adjusting based on the historical data of the current environment to adapt to the new environmental conditions, improving adaptability and detection accuracy.

[0132] For example, the fault threshold for a cable in a dry environment may be different from that in a humid environment. If the environment changes but a fixed threshold is still used, it may lead to misjudgment. With dynamic thresholds, the fault judgment criteria can be adjusted in real time according to the current environmental conditions, reducing false positives and missed positives.

[0133] The dynamic adjustment module is used to dynamically adjust the fault judgment threshold based on historical data.

[0134] Obtain the mean value of the operating status data of category h under the same historical environmental conditions as the current environmental conditions for similar groups. h , the standard deviation is σ h , and the preset optimization coefficient k h , and then substitute into the following formula:

[0135] T dynamic =μ h +k h ·σ h ;T dynamic Indicates the dynamic threshold of the running status corresponding to category h.

[0136] Many traditional high-voltage cable monitoring systems use fixed thresholds for fault diagnosis. For example, if the partial discharge signal exceeds a set fixed value, a fault is considered to exist. If the cable temperature exceeds a fixed threshold, an alarm is triggered. However, in actual applications, the working environment of high-voltage cables is changeable, and fixed thresholds are difficult to adapt to different environmental conditions, which may lead to false alarms or missed alarms: False alarms: Under environmental influences, some data exceeds the fixed threshold, but no fault actually occurs. Missed alarms: Due to different environments, some real faults are not detected by the fixed threshold. The operating status of the cable is affected by multiple environmental factors such as temperature, humidity, and electromagnetic interference. Under different conditions, the same fault characteristics may have different thresholds. By dynamically adjusting the threshold, the judgment criteria can be optimized based on historical data to improve detection accuracy.

[0137] The mean reflects the normal level of historical data under the same environmental conditions and serves as a benchmark for calculating dynamic thresholds. Avoid using fixed thresholds by directly assigning a single absolute value, which can result in inaccuracy in different environments. The standard deviation reflects the volatility of data under the same environmental conditions. A large standard deviation of data from a particular environment indicates significant data fluctuations in that environment, requiring a more flexible threshold for fault diagnosis. Adjusting the threshold based on the standard deviation can reduce false positives and false negatives. A large optimization coefficient means the system is more tolerant to environmental fluctuations, reducing false positives but potentially leading to false negatives. A small optimization coefficient means the system is more sensitive to abnormal changes, reducing false negatives but potentially leading to an increase in false positives. Adjusting the optimization coefficient based on historical data can optimize the system's detection performance and make it more suitable for specific application scenarios.

[0138] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0139] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0140] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A comprehensive online monitoring system for high-voltage cables in electrified railways, characterized in that: It includes data acquisition module, cluster analysis module, prediction method selection module, threshold determination module, dynamic adjustment module, and alarm module; The data acquisition module is used to obtain the environmental data and operating status data of the high-voltage cable; The cluster analysis module is used to perform cluster analysis based on the factory and usage environment data of high-voltage cables, and divide the cables into similar usage groups; The prediction method selection module is used to evaluate the stability and nonlinearity of environmental data, select the preset corresponding applicable prediction method based on the evaluation results, and predict the operating data group at the next moment; The threshold judgment module is used to predict the results and cable similarity groups, perform judgment analysis, and select fixed threshold or dynamic threshold method for fault judgment; The dynamic adjustment module is used to dynamically adjust the fault judgment threshold based on historical data to optimize the fault detection accuracy; The alarm module is used to trigger an early warning when the monitoring data exceeds a fixed threshold or a dynamic threshold.

2. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 1, characterized in that: Before cluster analysis, the factory data of high-voltage cables are converted into factory vectors, the usage environment data are converted into environment vectors, and the factory vectors and environment vectors are summarized into online monitoring vectors.

3. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 2, characterized in that: Cluster analysis uses the K-Means clustering algorithm or the DBSCAN clustering algorithm to cluster the current high-voltage cables with the rest of the high-voltage cables and divide the cables into similar groups.

4. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 3 is characterized in that: When evaluating based on the stability and nonlinearity of environmental data, a stability index for measuring the stability of the environmental data and a nonlinearity index for measuring the nonlinearity of the high-voltage cable are generated respectively.

5. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 4 is characterized in that: Selecting a preset applicable prediction method based on the evaluation results and predicting the next moment's operating data group means: A pre-trained machine learning model, namely the convolutional neural network model, is used. The stability index and the nonlinear index are used as input data, and the type of prediction method is used as output data. The prediction methods include time series prediction method and regression prediction method.

6. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 5, characterized in that: The logic for obtaining the stability index is: Calculate the standard deviation of one type of environmental data within a preset time window: X i is the data value corresponding to the sampling point i of one type of environmental data in the preset time window, μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, and σ is the standard deviation of one type of environmental data in the preset time window; Calculate the coefficient of variation of the data to normalize the degree of data fluctuation: CV is the coefficient of variation of the data; The calculation formula of the stability coefficient is: SI=e -α·CV ; α is the preset non-zero adjustment coefficient, SI is the stability coefficient, and among all categories of environmental data, the maximum value of the stability coefficient is taken as the stability index.

7. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 6, characterized in that: The logic for obtaining the nonlinear index is: Get the first-order difference sequence of one type of environmental data within the preset time window, that is, the rate of change: ΔX i =X i+1 -X i ;X i is the data value corresponding to the sampling point i of one type of environmental data within the preset time window, X i+1 is the data value corresponding to the sampling point i+1 of one type of environmental data within the preset time window, ΔX i The change rate of sampling point i+1 of one type of environmental data relative to sampling point i within the preset time window; Calculate the coefficient of variation of the difference: σΔX and μΔX are the standard deviation and mean of the first-order difference sequence, CV Δ is the coefficient of variation of the difference; Calculate the autocorrelation coefficient: μ is the mean of one type of environmental data in the preset time window, N is the number of sampling times in the preset time window, k is the preset time lag order, X i+k is the data value corresponding to the sampling point i+k of one type of environmental data within the preset time window, and ACF(k) is the autocorrelation coefficient of one type of environmental data within the preset time window; The nonlinear coefficient calculation formula is: NLI=β1·CV Δ +β2·(1-|ACF(1)|); β1 and β2 are both preset non-zero proportional coefficients, and their sum is one. ACF(1) is the autocorrelation coefficient corresponding to the value of k equal to 1. NLI is the nonlinear coefficient. Among all categories of environmental data, the maximum value of the nonlinear coefficient is taken as the nonlinear index.

8. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 7, characterized in that: The threshold judgment module is used to predict the results and cable similarity groups, perform judgment analysis, and select fixed threshold or dynamic threshold method for fault judgment. The current environmental data group composed of the prediction results of multiple types of environmental data is compared with the group environmental data group of similar cable groups under current environmental conditions, and the Euclidean distance between the current environmental data group and the group environmental data group is calculated. If the Euclidean distance is greater than the preset deviation threshold, the dynamic threshold method is used for fault judgment. If the Euclidean distance is less than or equal to the preset deviation threshold, the fixed threshold method is used for fault judgment.

9. The comprehensive online monitoring system for high-voltage cables of electrified railways according to claim 8, characterized in that: The dynamic adjustment module is used to dynamically adjust the fault judgment threshold based on historical data. Obtain the mean value of the operating status data of category h under the same historical environmental conditions as the current environmental conditions for similar groups respectively. h , the standard deviation is σ h , and the preset optimization coefficient k h , and then substitute into the following formula: T dynamic =μ h +k h ·σ h ;T dynamic Indicates the dynamic threshold of the operating state corresponding to category h.

Images