High-speed rail steel rail insulation joint health monitoring and early warning method and system

Abstract

The invention provides a high-speed rail insulation joint health monitoring and early warning method and system, and relates to the technical field of joint health monitoring, and the method comprises the steps: carrying out the dimension reduction of resistance data and environmental influence factors through a random projection algorithm, carrying out the anomaly detection of the resistance feature data after the dimension reduction through an isolated forest algorithm, and carrying out the early warning. And through weight adjustment and path length optimization, a resistance abnormal point is marked. Using an empirical mode decomposition method to carry out constrained decomposition on the abnormal point data, and extracting multi-scale intrinsic mode function feature data; and performing resistance change trend prediction by using a Gaussian process regression model to obtain a resistance change prediction value and a confidence interval thereof. And inputting the predicted value and the confidence interval into a fuzzy reasoning model to obtain a graded early warning result of the high-speed rail insulated joint. Through multi-level data processing and analysis, the accuracy and reliability of fault early warning are improved, and safe operation of a high-speed rail steel rail system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of joint health monitoring, and in particular, to a method and system for health monitoring and early warning of high-speed rail rail insulation joints. Background Art

[0002] At present, high-speed rail rail insulation joints are key components in the high-speed rail track circuit system, and their health status is crucial for the safety and stability of high-speed rail operation. The health status of rail insulation joints is usually evaluated by monitoring the change of resistance value, and the resistance value is affected by various environmental factors such as temperature, humidity, and vibration, and is prone to abnormal fluctuations. Existing monitoring methods usually detect abnormal points based on fixed threshold judgment or simple statistical analysis techniques. However, these methods have limited ability to process multi-source data in complex environments, are easily affected by noise interference, and have insufficient sensitivity to early abnormalities, resulting in inaccurate early warning results. And the existing technology cannot provide accurate and real-time health status evaluation and fault prediction.

[0003] Therefore, there is an urgent need for a method and system for health monitoring and early warning of high-speed rail rail insulation joints to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for health monitoring and early warning of high-speed rail rail insulation joints to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] In a first aspect, the present application provides a method for health monitoring and early warning of high-speed rail rail insulation joints, including:

[0006] Obtain the resistance data of the rail insulation joint and environmental impact factors, where the environmental impact factors include temperature data, humidity data, and vibration data of the rail insulation joint;

[0007] Perform dimensionality reduction processing on the resistance data of the rail insulation joint and environmental impact factors based on the random projection algorithm to obtain the dimensionality-reduced resistance feature data;

[0008] Perform anomaly detection on the dimensionality-reduced resistance feature data based on the isolation forest algorithm, where, through weight adjustment and dynamic optimization of the path length, obtain the labeled data containing resistance anomaly points;

[0009] Decompose the labeled data containing resistance anomaly points based on the empirical mode decomposition method, where, through constrained decomposition and instantaneous feature extraction of the labeled data containing resistance anomaly points, obtain the multi-scale intrinsic mode function feature data;

[0010] Send the multi-scale intrinsic mode function feature data to a preset Gaussian process regression model for trend prediction. Among them, through kernel function combination and hyperparameter adjustment, the predicted value of the resistance change and the corresponding confidence interval are obtained;

[0011] Send the predicted value of the resistance change and the corresponding confidence interval to a preset fuzzy inference model for fuzzy inference. Among them, through dynamically adjusting the weights of the fuzzy rules and judging the confidence level, the hierarchical warning result of the high-speed rail rail insulation joint is obtained.

[0012] In a second aspect, the present application also provides a health monitoring and warning system for high-speed rail rail insulation joints, including:

[0013] An acquisition unit, configured to acquire the resistance data of the rail insulation joint and environmental impact factors, where the environmental impact factors include temperature data, humidity data, and vibration data of the rail insulation joint;

[0014] A processing unit, configured to perform dimensionality reduction processing on the resistance data of the rail insulation joint and environmental impact factors based on a random projection algorithm to obtain the dimensionality-reduced resistance feature data;

[0015] A detection unit, configured to perform anomaly detection on the dimensionality-reduced resistance feature data based on an isolation forest algorithm. Among them, through weight adjustment and dynamic optimization of the path length, the labeled data including resistance anomaly points is obtained;

[0016] An analysis unit, configured to decompose the labeled data including resistance anomaly points based on an empirical mode decomposition method. Among them, through constrained decomposition and instantaneous feature extraction of the labeled data including resistance anomaly points, multi-scale intrinsic mode function feature data is obtained;

[0017] A prediction unit, configured to send the multi-scale intrinsic mode function feature data to a preset Gaussian process regression model for trend prediction. Among them, through kernel function combination and hyperparameter adjustment, the predicted value of the resistance change and the corresponding confidence interval are obtained;

[0018] A judgment unit, configured to send the predicted value of the resistance change and the corresponding confidence interval to a preset fuzzy inference model for fuzzy inference. Among them, through dynamically adjusting the weights of the fuzzy rules and judging the confidence level, the hierarchical warning result of the high-speed rail rail insulation joint is obtained.

[0019] The beneficial effects of the present invention are:

[0020] The present invention obtains resistance data and environmental factor data, performs dimensionality reduction processing on resistance feature data based on the random projection algorithm, combines the isolation forest algorithm for anomaly detection, dynamically optimizes weights and path lengths, and realizes accurate anomaly point marking; further adopts the empirical mode decomposition method to perform multi-scale analysis on the marked data, extracts instantaneous features and generates multi-scale intrinsic mode function feature data; based on the Gaussian process regression model, performs trend prediction on the multi-scale feature data, and generates the predicted value and confidence interval of the resistance change through kernel function combination and hyperparameter optimization; finally, dynamically adjusts the rule weights and judgment confidence through the fuzzy inference model, and outputs a hierarchical early warning result. The present invention effectively solves the deficiencies in the prior art of insufficient processing of multi-source heterogeneous data, low anomaly detection accuracy, lack of trend prediction and hierarchical early warning, and significantly improves the accuracy, robustness and real-time performance of the health monitoring of rail insulation joints.

[0021] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification, or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is a schematic flow chart of the health monitoring and early warning method for high-speed rail rail insulation joints described in the embodiments of the present invention;

[0024] Figure 2 It is a schematic structural diagram of the health monitoring and early warning system for high-speed rail rail insulation joints described in the embodiments of the present invention.

[0025] In the figure: 701, acquisition unit; 702, processing unit; 703, detection unit; 704, analysis unit; 705, prediction unit; 706, judgment unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, terms such as "first" and "second" are only used for descriptive distinction and cannot be understood as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a method for health monitoring and early warning of high-speed rail rail insulation joints.

[0030] See Figure 1 , which shows that this method includes step S1, step S2, step S3, step S4, step S5, and step S6.

[0031] Step S1: Obtain the resistance data of the rail insulation joint and environmental impact factors, where the environmental impact factors include the temperature data, humidity data, and vibration data of the rail insulation joint;

[0032] It can be understood that the acquisition of resistance data in this step usually relies on high-precision resistance measurement equipment, and its measurement interval and sampling frequency are optimized according to the actual needs of high-speed rail operation to ensure the timeliness and accuracy of key data. The acquisition of environmental factors includes temperature, humidity, and vibration data. Among them, the temperature data is obtained through an embedded thermistor sensor, the humidity data is captured by a highly sensitive capacitive humidity sensor, and the vibration data is captured by an accelerometer sensor to capture the dynamic response of the rail under high-frequency and low-frequency vibration conditions. By accurately acquiring and real-time transmitting high-quality multi-source data, the reliability of the monitoring system is laid, and at the same time, the detection errors caused by data loss or inaccuracy are reduced, providing a complete input data set for subsequent anomaly detection and trend prediction.

[0033] Step S2: Perform dimensionality reduction on the resistance data of the rail insulation joint and environmental impact factors based on the random projection algorithm to obtain the dimensionality-reduced resistance feature data;

[0034] It can be understood that this step significantly reduces the redundancy and computational complexity of high-dimensional data, while maintaining the main features of the data, improving the accuracy and efficiency of subsequent anomaly detection. In addition, by introducing weighted reconstruction, it is ensured that the dimensionality-reduced feature data is more in line with the actual working scenario of the rail insulation joint, making it more sensitive to environmental changes and enhancing the robustness and accuracy of anomaly point detection. In this step, step S2 includes steps S21, S22, S23, and S24.

[0035] Step S21: Standardize the resistance data of the rail insulation joint and environmental impact factors. Among them, standardize the resistance data of the rail insulation joint and environmental impact factors through the Z-score standardization method to obtain the standardized joint feature data of high-dimensional resistance and environmental factors;

[0036] It can be understood that this step preprocesses the resistance data of the rail insulation joint and environmental impact factors through the Z-score standardization method to generate the standardized joint feature data of high-dimensional resistance and environmental factors. Since the numerical ranges and distribution characteristics of the resistance data of the rail insulation joint and environmental impact factors (such as temperature, humidity, and vibration) are different, directly modeling them may cause some features to be overestimated or underestimated due to the influence of dimensionality differences or extreme values. Through standardization, the characteristic values of the resistance data and environmental factors are within the same scale range, ensuring the numerical stability and convergence in the subsequent random projection and dimensionality reduction processes.

[0037] Step S22: Generate a random Gaussian matrix based on the above joint feature data. Among them, randomly generate a sparse matrix through Gaussian distribution to map the feature vectors of the original high-dimensional space to obtain a random mapping matrix;

[0038] It can be understood that each element of the random Gaussian matrix in this step is randomly generated according to an independent and identically distributed Gaussian distribution. Sparsity is achieved by introducing specific probability rules when generating the matrix. For example, a sparsification strategy is used to set some elements of the matrix to zero to reduce the computational complexity and improve efficiency. This construction method conforms to the Johnson-Lindenstrauss lemma, which states that by appropriately choosing a random projection matrix, the Euclidean distance between high-dimensional features can be approximately preserved in a low-dimensional space. In addition, there is a certain correlation between the resistance characteristics of high-speed rail track insulation joints and environmental influence factors. The sparsification process of the random mapping matrix can reduce the interference of correlated features on the mapping result and improve the computational efficiency at the same time. Once the random mapping matrix is generated, it can be used as a fixed mapping operator to perform a linear transformation on the standardized joint feature data to obtain a random projection result suitable for dimensionality reduction processing.

[0039] Step S23: Perform feature mapping processing on the random mapping matrix based on the random projection algorithm, map the high-dimensional feature vector to a low-dimensional space, and obtain the resistance feature data after preliminary dimensionality reduction.

[0040] It can be understood that the random projection algorithm is used to perform feature mapping processing on the generated random Gaussian matrix, map the high-dimensional feature vector to a low-dimensional space, and generate the resistance feature data after preliminary dimensionality reduction. The uniqueness of this mapping process in the monitoring scenario of high-speed rail track insulation joints lies in that the dimensions of the original resistance data and environmental factor data are relatively high and there is noise and redundancy. The random projection algorithm avoids the dependence on the covariance matrix decomposition of traditional dimensionality reduction methods (such as principal component analysis) through a global random mapping operation, thus avoiding the computational overhead caused by the complexity of the feature distribution. At the same time, its unsupervised nature allows for efficient processing of unlabeled feature data.

[0041] Step S24: Perform feature importance reconstruction processing on the resistance feature data after preliminary dimensionality reduction, and optimize and adjust the resistance feature data after preliminary dimensionality reduction through the weighted principal component reconstruction algorithm to obtain the resistance feature data after final dimensionality reduction.

[0042] It is understandable that this step first conducts a feature importance analysis on the preliminarily dimension-reduced data. The methods used include calculating the contribution of each feature to the overall variance in the data distribution and its correlation with the target analysis task (such as anomaly detection). For the high-speed rail rail insulation joint scenario, changes in environmental factors (temperature, humidity, vibration) may affect the variation law of resistance, so these features are given higher weights. Among them, feature importance is quantified through indicators such as variance contribution rate, mutual information, or correlation coefficient. Based on the above feature importance information, weighted principal component analysis is used to reconstruct and optimize the preliminarily dimension-reduced data. And feature weights are introduced in the calculation of the feature covariance matrix of principal component analysis, and the formula is as follows:

[0043] C′ = W·C·W T

[0044] Among them, C is the covariance matrix of the original preliminarily dimension-reduced data, W is the feature weight matrix, and C′ is the optimized covariance matrix.

[0045] Step S3: Perform anomaly detection on the dimension-reduced resistance feature data based on the isolation forest algorithm. Among them, through weight adjustment and dynamic optimization of the path length, marked data containing resistance anomaly points are obtained;

[0046] It is understandable that this step analyzes the dimension contribution degree of the dimension-reduced resistance feature data, calculates the role size of each feature in the overall anomaly detection, and assigns weights based on feature importance. This weight adjustment mechanism ensures greater sensitivity to anomaly points of key features. For example, features that may significantly affect the health status of rail insulation joints. Subsequently, by dynamically optimizing the calculation method of the path length, combined with the distribution characteristics of the resistance feature data, the splitting depth and node coverage are adjusted, thereby reducing the false positive rate in sparse regions. In this step, step S3 includes step S31, step S32, step S33, and step S34.

[0047] Step S31: Initialize the dimension-reduced resistance feature data. Among them, by constructing a set of decision trees based on random samples and defining an anomaly scoring formula, a basic isolation forest model is obtained;

[0048] It is understandable that first, a sub-sample set is randomly selected from the dimension-reduced resistance feature data to reduce the computational complexity and avoid the excessive influence of noise points. Based on the sub-samples, multiple isolation trees are constructed using randomly selected feature dimensions and randomly divided thresholds. Each isolation tree corresponds to a set of random splitting rules, and the data points are gradually separated to the leaf nodes through recursive partitioning. The depth of the tree reflects the difficulty of isolating the data points. Generally, outliers can be isolated at a shallower level because they are located in the sparse region of the data distribution. Next, an outlier scoring formula for the isolation forest is defined, and the outlier scoring formula is as follows:

[0049]

[0050] Among them, h(x) represents the theoretical path length of the isolation tree, E(h) represents the expected path length, and s(x) represents the outlier score.

[0051] Step S32: Adjust the weights of the basic isolation forest model. Among them, by calculating the contribution degree of the resistance feature dimension and dynamically allocating weights, a weighted isolation forest model is obtained;

[0052] It is understandable that first, it is necessary to calculate the contribution degree of each resistance feature dimension to the overall model. The contribution degree can be evaluated by quantifying the influence of the feature when splitting nodes in the isolation forest. Generally, the contribution degree of a feature is related to the splitting frequency of the feature in the isolation tree and the degree of influence on the path length. For example, if a certain dimension frequently appears as a splitting point in multiple trees, and this dimension can significantly reduce the path length of the tree, it indicates that this feature has a high importance for distinguishing normal and abnormal data points, and its contribution degree is the division of the splitting frequency by the path length. Then, based on the contribution degree of each feature, the weights of each feature in the isolation forest model are adjusted through a weighting strategy. Under this weighting mechanism, features with high contribution degrees will have larger weights, enabling them to play a more dominant role in the tree construction process. For features with low contribution degrees, their weights are appropriately reduced to reduce their negative impact on the model performance. This process is actually dynamically adjusting the influence of features during the training process to help the model better adapt to the key information in the data. Finally, a weighted isolation forest model is obtained.

[0053] Step S33: Perform dynamic optimization processing on the path length of the weighted isolation forest model. Among them, by adjusting the splitting depth and the node coverage area, an optimized outlier scoring model is obtained;

[0054] It can be understood that in the Isolation Forest, data points are gradually "isolated" until a termination condition is reached (e.g., a certain splitting threshold or maximum depth). By dynamically adjusting the splitting depth, the structure of the decision tree can be optimized to more accurately reflect the distribution characteristics of the data. For example, in certain data point distribution areas, deeper splitting may be required to distinguish normal data from abnormal data, while in areas where the data is more concentrated, the splitting depth can be appropriately reduced. In this way, the calculation of the path length can more accurately reflect the abnormality of the data. And each tree in the Isolation Forest constructs nodes by randomly selecting features and data subsets. The adjustment of the node coverage area refers to changing the range of node partitioning so that some more important features can more effectively distinguish the data. For abnormal data, it may be necessary to narrow the coverage area to improve the model's sensitivity to identifying abnormal points. On the other hand, for normal data, appropriately expanding the node coverage area can reduce the possibility of model misjudgment. Through these adjustments, the optimized model can more accurately evaluate the anomaly score of each data point. Especially when dealing with complex rail insulation joint resistance data, environmental impact factors such as temperature and humidity fluctuations may cause changes in the resistance data, and these changes may be similar to the characteristics of abnormal points. By dynamically optimizing the path length, the model can better eliminate the interference of environmental noise and accurately identify abnormal changes in the resistance.

[0055] Step S34: Based on the optimized anomaly scoring model, score and perform threshold partitioning on all the dimensionality-reduced resistance feature data to obtain labeled data containing resistance abnormal points.

[0056] It can be understood that by using the Isolation Forest algorithm to model the dimensionality-reduced resistance feature data, the anomaly score of each data point is obtained. This score reflects the degree to which the data point is isolated. Generally speaking, abnormal points have higher scores because they are more likely to be "isolated" on the decision tree path. The score value is usually a continuous numerical value indicating the likelihood that a certain data point is a normal data point or an abnormal data point. According to the distribution of the scores, a threshold is set to determine which data points are considered abnormal. This threshold is usually determined through experience or based on a certain statistical analysis method. In this step, by observing the distribution of the scores, a suitable quantile is selected as the threshold. Through the optimized anomaly scoring model and reasonable threshold partitioning, abnormal points in the resistance data can be accurately identified. This is particularly important for the monitoring of high-speed rail rail insulation joints because abnormal changes in the resistance may be early signs of joint failures. By promptly identifying these abnormal points, maintenance measures can be taken in advance to avoid safety accidents caused by failures and improve the operation stability and maintenance efficiency of the system.

[0057] Step S4: Decompose the marked data containing resistance anomaly points using the empirical mode decomposition method. Specifically, through constrained decomposition and instantaneous feature extraction of the marked data containing resistance anomaly points, multi-scale intrinsic mode function feature data is obtained;

[0058] It can be understood that through empirical mode decomposition and instantaneous feature extraction, the hidden change patterns in resistance data, especially the features related to resistance anomaly points, can be deeply explored. Constrained decomposition and multi-scale analysis further improve the accuracy of the decomposition results, making the anomaly point features more prominent and facilitating subsequent trend prediction and early warning. This process significantly enhances the ability to analyze resistance data, providing more accurate and multi-dimensional information for the health monitoring and fault early warning of high-speed rail rail insulation joints, enabling the early detection of potential fault risks and ensuring the safe operation of the system. In this step, step S4 includes steps S41, S42, S43, and S44.

[0059] Step S41: Perform initial empirical mode decomposition on the marked data containing resistance anomaly points. Specifically, through the constrained decomposition strategy of extracting intrinsic mode functions, a set of initially decomposed intrinsic mode functions is obtained;

[0060] It can be understood that in this step, the initial empirical mode decomposition method is an adaptive signal decomposition technique that can decompose the original non-linear and non-stationary signal into a series of intrinsic mode functions. Each intrinsic mode function represents a specific frequency component of the signal. For resistance data, the initial empirical mode decomposition decomposes the signal into multiple intrinsic mode functions, reflecting the dynamic characteristics of the resistance value at different time scales. Here, the resistance data is used as the input signal, and after the decomposition process of the initial empirical mode decomposition, multiple intrinsic mode functions are obtained. Among them, the constraint conditions include: Frequency constraint: Restrict the frequency range of each intrinsic mode function so that each obtained intrinsic mode function corresponds to a reasonable change period. Stationarity constraint: By controlling the stationarity of the signal, unnecessary frequency fluctuations or pseudo-modes are avoided. Amplitude constraint: For abnormal signals, control the amplitude range of the intrinsic mode function to avoid excessive abnormal values or irrelevant fluctuations being extracted as effective modes.

[0061] Step S42: Perform noise suppression on the set of initially decomposed intrinsic mode functions. Specifically, remove the intrinsic mode function components with low signal-to-noise ratio through singular value decomposition to obtain a set of denoised intrinsic mode functions;

[0062] It can be understood that in this step, each intrinsic mode function is regarded as a matrix, and singular value decomposition is performed. The time series of each intrinsic mode function can be represented as a vector, and after forming a matrix, singular value decomposition is carried out. In this way, we can obtain the singular values and corresponding singular vectors of the intrinsic mode function signal. Among the results of the singular values, the singular values represent the "importance" or energy magnitude of the signal. Intrinsic mode functions with low signal-to-noise ratio usually correspond to smaller singular values, indicating that they carry less energy and are more likely to be noise. Through the singular value decomposition method, the noise components that may be generated during the decomposition of the resistance signal can be effectively removed. By removing the intrinsic mode function components with low signal-to-noise ratio, it can be ensured that the signal features used in subsequent analysis are more pure and stable, thereby improving the accuracy of subsequent resistance anomaly detection and trend prediction. In addition, this noise suppression process can significantly improve the sensitivity of the system to changes in resistance anomaly points, making the subsequent multi-scale feature extraction and fuzzy inference more reliable and physically meaningful.

[0063] Step S43: Analyze the instantaneous frequency and amplitude of all intrinsic mode function components through Hilbert transform, and construct an instantaneous feature matrix.

[0064] It can be understood that in this step, the instantaneous frequency and amplitude of each intrinsic mode function component are analyzed through the above-mentioned Hilbert transform. At this time, the instantaneous frequency and amplitude of each intrinsic mode function can be regarded as sequences that change with time, and these sequences represent the dynamic characteristics of the signal. Organize the instantaneous frequency and amplitude data of all intrinsic mode functions into a matrix, where each row represents the instantaneous features at a time point, and each column represents the instantaneous frequency or amplitude of an intrinsic mode function component. This matrix is the "instantaneous feature matrix", which aggregates the time-frequency characteristics of all intrinsic mode function components and reflects the variation laws of the signal at multiple scales.

[0065] Step S44: Perform frequency band decomposition and reconstruction on the instantaneous features in the instantaneous feature matrix through wavelet packet transform to obtain multi-scale intrinsic mode function feature data.

[0066] It can be understood that the wavelet packet transform in this step is a signal processing technique that recursively decomposes the spectrum of a signal to provide a finer frequency band decomposition than the traditional wavelet transform. Different from the wavelet transform that only decomposes the low-frequency part, the wavelet packet transform further decomposes each layer of details of the signal, thereby obtaining a multi-level and full-frequency band frequency domain representation. In the wavelet packet transform, the signal is first decomposed into low-frequency and high-frequency components, and then both the low-frequency part and the high-frequency part will continue to be decomposed. This multi-level decomposition method enables the wavelet packet transform to analyze different scales of the signal with a higher frequency resolution. The wavelet packet transform can effectively separate the characteristics of different frequency bands in the signal, and then analyze different aspects of the resistance change from multiple scales. This analysis method provides a finer characterization of the complexity of the resistance signal and can better adapt to the changes of different scales in the signal.

[0067] Step S5: Send the multi-scale intrinsic mode function feature data to a preset Gaussian process regression model for trend prediction. Among them, through the combination of kernel functions and hyperparameter adjustment, the predicted value of the resistance change and the corresponding confidence interval are obtained;

[0068] It can be understood that by combining the kernel function combination and hyperparameter optimization, the complex patterns in the resistance data can be flexibly fitted, and the non-linear and multi-level change trends can be captured. This provides a powerful modeling ability for the prediction of the resistance change trend. By providing the confidence interval, the user can not only obtain a point prediction value, but also evaluate the uncertainty of the prediction. This information is particularly important for the decision-making process, especially in the health monitoring of high-speed rail rail insulation joints, which can help predict potential failure risks and take timely measures. In this step, step S5 includes step S51, step S52, step S53 and step S54.

[0069] Step S51: Perform kernel mapping processing on the multi-scale intrinsic mode function feature data. Among them, through the mapping of the high-dimensional feature space based on the combination of kernel functions, the high-dimensional feature data after kernel mapping is obtained;

[0070] It can be understood that in this step, a kernel function combination is used for feature mapping. A kernel function is a function used to measure the similarity between data points, and its selection is crucial for the mapping effect. Through the kernel function combination, the system can perform diverse mappings according to different characteristics of the data. In this step, for data with complex non-linear relationships, a radial basis function kernel is adopted, while for relatively linear data, a linear kernel function is used. The kernel function combination can improve the mapping quality by adapting to the diversity of the data, enabling the data in the high-dimensional feature space to better express the resistance change trend. After the mapping by the kernel function combination, the low-dimensional intrinsic mode function feature data is projected into a higher-dimensional feature space. In this new space, the relationships between the data become more obvious and distinguishable, providing richer information for subsequent regression analysis.

[0071] Step S52: Send the high-dimensional feature data after kernel mapping to a preset Gaussian process regression model for training, where the kernel matrix of the high-dimensional feature data after kernel mapping is calculated and the hyperparameters are optimized to obtain an optimized Gaussian process model;

[0072] It can be understood that in this step, before passing the high-dimensional feature data after kernel mapping to the Gaussian process regression model, it is necessary to calculate the similarity between data points, which is usually achieved by calculating the kernel matrix. Each element of the kernel matrix represents the similarity or correlation between two data points. The performance in the Gaussian process model strongly depends on the choice of the kernel function and the corresponding hyperparameters (such as length scale, noise variance, etc.). These hyperparameters determine the shape of the kernel function and the flexibility of the model, so an optimization process is required to select the hyperparameters most suitable for the data. The optimization of the hyperparameters in this step is completed by maximizing the log-likelihood function of the model, that is, by adjusting the hyperparameters to minimize the prediction error of the model on the training data. The optimization methods in this step are gradient descent method, Bayesian optimization, etc.

[0073] Step S53: Use the optimized Gaussian process model to predict the trend of resistance change, where the predicted value of the resistance change is obtained by calculating the posterior distribution;

[0074] It can be understood that in this step, the prediction of the model depends on the posterior distribution, and the calculation formula of its posterior distribution is as follows:

[0075]

[0076] where, μ * represents the mean of the predicted value, ∑ * represents the covariance of the predicted value, x * represents the new input point, y represents the observed data, X represents the input data, and y * represents the prediction result obtained by calculating the posterior distribution of the Gaussian process, Represents a normal distribution.

[0077] Step S54: Dynamically optimize the confidence interval of the predicted value of the resistance change to obtain a prediction result with anomaly marks and the corresponding confidence interval.

[0078] It can be understood that in this step, in addition to the trend prediction value, the posterior covariance also provides a measure of the uncertainty of the prediction value. The standard deviation represents the confidence level of the prediction. Generally, in a regression model, the confidence interval is determined based on the mean plus or minus a certain range of standard deviations. For example, the following formula can be used to generate the confidence interval:

[0079]

[0080] where A represents the confidence interval, μ * represents the mean of the predicted value, κ is a constant related to the confidence level, and ∑ * is the covariance matrix of the prediction.

[0081] Step S6: Send the predicted value of the resistance change and the corresponding confidence interval to a preset fuzzy inference model for fuzzy inference, where a hierarchical warning result of the high-speed rail rail insulation joint is obtained by dynamically adjusting the weights of the fuzzy rules and judging the confidence level.

[0082] It can be understood that by taking the resistance change prediction and the confidence interval together as inputs, the fuzzy inference model not only processes the trend of the resistance change but also considers the uncertainty of the prediction, thus providing a decision-making support under uncertain conditions. By predicting in advance and issuing hierarchical alarms, faults can be effectively prevented, maintenance costs can be reduced, and the safety and reliability of high-speed rail operation can be improved. In this step, Step S6 includes Step S61, Step S62, Step S63, and Step S64.

[0083] Step S61: Initialize the predicted value of the resistance change and the corresponding confidence interval, where a fuzzy membership function is constructed based on the resistance change characteristics and the initial fuzzy rules are determined to obtain an initial fuzzy rule set and the corresponding membership function;

[0084] It can be understood that in this step, by constructing a fuzzy membership function based on the resistance change characteristics, the system can convert the numerical value of the resistance change into a fuzzy value for further reasoning and judgment. When constructing the fuzzy membership function, the different amplitudes of the resistance change and their impacts on the health status of the rail insulating joint are usually considered. For example, the resistance change amount is mapped into fuzzy categories such as "normal", "abnormal", or "severely abnormal". This conversion can transform traditional quantitative data into information with uncertainty and ambiguity that can be processed, facilitating subsequent fuzzy reasoning. Fuzzy rules are usually set according to the resistance change pattern, historical data, and expert knowledge, etc. For example, if the resistance change value is large and the confidence interval is wide, it may indicate that the system is approaching a fault and a higher warning level is required. Fuzzy rules are set not only based on the actual numerical change of the resistance but also considering the trend and uncertainty of the resistance change to ensure the adaptability and accuracy of the system under different working conditions.

[0085] In addition, during the initialization process, it is also necessary to determine the initial fuzzy rule set. Fuzzy rules are usually set according to the resistance change pattern, historical data, and expert knowledge, etc. For example, if the resistance change value is large and the confidence interval is wide, it may indicate that the system is approaching a fault and a higher warning level is required. Fuzzy rules need to be set not only based on the actual numerical change of the resistance but also considering the trend and uncertainty of the resistance change to ensure the adaptability and accuracy of the system under different working conditions. Among them, in this step, a trapezoidal membership function is used to represent different resistance change levels such as "normal", "abnormal", and "severely abnormal". For example: if the resistance change amount is small and the confidence interval is narrow, the membership function can be used to represent the "normal" state. When the resistance change amount is large and the confidence interval is wide, the membership function can represent the "abnormal" or "severely abnormal" state.

[0086] Step S62: Based on the weighted fuzzy inference mechanism, assign dynamic weights to the initial fuzzy rule set and the corresponding membership function to obtain an optimized fuzzy rule set with dynamic weights.

[0087] It can be understood that the initial fuzzy rule set is constructed through a fuzzification process by the membership functions corresponding to the predicted values of resistance changes and confidence intervals. These rules are usually formulated based on domain knowledge or expert experience. For example, "small resistance change and narrow confidence interval" may correspond to the "normal" state, while "large resistance change and wide confidence interval" may correspond to the "abnormal" or "severely abnormal" state. At this time, each fuzzy rule has a preliminary membership function to represent its corresponding resistance change state. Then, dynamic weight assignment is achieved by introducing a weighting mechanism, making the contribution degrees of different rules to the final judgment result different. The basis for weight assignment can be the importance of the rule, the reliability of the data on which the rule is based, or the weights obtained through a learning algorithm. The specific method is to adjust the weights of each rule according to the actual distribution of resistance data and confidence intervals. For example, if the resistance changes frequently or the change trend is more obvious in certain regions, the weights of these rules can be increased to make their influence greater during reasoning.

[0088] Step S63: Perform fuzzy reasoning based on the fuzzy rule set optimized by dynamic weights. Among them, the fuzzy reasoning process includes fuzzifying the fuzzy input, performing rule reasoning, and defuzzifying the fuzzy output to obtain the fuzzy reasoning result of the resistance change.

[0089] It can be understood that in this step, the fuzzification of the fuzzy input: Fuzzification is the process of converting the actual predicted values of resistance changes and confidence intervals into fuzzy sets. The original numerical inputs (such as resistance values, predicted values, and their confidence intervals) are usually specific quantitative data. However, in fuzzy reasoning, these data need to be converted into "fuzzy membership degrees" of different levels. For example, the predicted value of the resistance change may be mapped to three fuzzy subsets: "low change", "medium change", or "high change", and each category is assigned a corresponding membership degree (a value between 0 and 1), indicating the degree to which the input value belongs to this fuzzy category. At the same time, the width of the confidence interval also affects the distribution of the membership degree. A wider confidence interval may reduce the membership degree of a certain rule.

[0090] The rule reasoning process: Rule reasoning is a process of reasoning based on the previously constructed fuzzy rule set and its dynamic weights. Each rule evaluates the fuzzy membership degree of the input and adjusts its contribution to the reasoning result according to the dynamic weight. During the rule reasoning process, if the predicted value of the resistance change and the confidence interval meet the conditions of a certain fuzzy rule, then this rule will be triggered. The triggered rule is calculated according to its corresponding weight to generate a fuzzy reasoning result. For example, a certain rule may judge whether to trigger the "high risk" state based on the membership degree of the input and adjust the risk level according to its weight.

[0091] Defuzzification of fuzzy output: Defuzzification is the process of converting the result of fuzzy inference into an actual numerical output. The result of fuzzy inference is usually a fuzzy set, such as the "high risk" state, which needs to be converted into a specific numerical value (such as the risk level or risk value of resistance change) through defuzzification methods. The defuzzification method in this step is the weighted average method. The weighted average method obtains the final numerical output by weighted summation of all fuzzy results. In this step, the defuzzification process will provide a specific risk level or warning value for the prediction of resistance change, indicating whether further monitoring or maintenance is required.

[0092] Step S64: Perform hierarchical early warning processing on the fuzzy inference result of the resistance change according to a preset risk level threshold of resistance change.

[0093] It can be understood that in this step, a set of clear risk level thresholds of resistance change need to be preset according to the characteristics of resistance change and its impact on the health of rail insulation joints. These thresholds are based on empirical data or historical fault data and usually include multiple levels, such as "normal", "warning", "severe", "critical", etc. The specific risk level thresholds depend on factors such as the amplitude, trend, and width of the confidence interval of resistance change. For each type of resistance change prediction, the system will map it to a specific risk level according to the range of its change value. For example, if the resistance change exceeds a certain threshold and the confidence interval is narrow, it may trigger a "severe" or "critical" level early warning. If the predicted value of resistance change and the confidence interval indicate that the risk at this point is at a certain specific level (such as "high risk"), then this point will be marked as an early warning of this level. This process is based on the defined risk level thresholds and corresponding rules to ensure that different degrees of risks are reasonably responded to. For example, if the resistance change is at the "warning" level, the system may issue maintenance suggestions; while if the change is at the "critical" level, it may trigger emergency inspection and repair measures.

[0094] Embodiment 2:

[0095] As Figure 2 shown, this embodiment provides a health monitoring and early warning system for high-speed rail rail insulation joints. Refer to Figure 2 The system includes an acquisition unit 701, a processing unit 702, a detection unit 703, an analysis unit 704, a prediction unit 705, and a judgment unit 706.

[0096] The acquisition unit 701 is used to acquire the resistance data of the rail insulation joint and environmental impact factors, and the environmental impact factors include temperature data, humidity data, and vibration data of the rail insulation joint;

[0097] A processing unit 702, configured to perform dimensionality reduction processing on the resistance data of the rail insulation joint and environmental impact factors based on a random projection algorithm to obtain the dimensionality-reduced resistance feature data;

[0098] A detection unit 703, configured to perform anomaly detection on the dimensionality-reduced resistance feature data based on an isolation forest algorithm, wherein, through weight adjustment and path length dynamic optimization, marked data including resistance anomaly points is obtained;

[0099] An analysis unit 704, configured to decompose the marked data including resistance anomaly points based on an empirical mode decomposition method, wherein, through constrained decomposition and instantaneous feature extraction of the marked data including resistance anomaly points, multi-scale intrinsic mode function feature data is obtained;

[0100] A prediction unit 705, configured to send the multi-scale intrinsic mode function feature data to a preset Gaussian process regression model for trend prediction, wherein, through kernel function combination and hyperparameter adjustment, a predicted value of the resistance change and the corresponding confidence interval are obtained;

[0101] A judgment unit 706, configured to send the predicted value of the resistance change and the corresponding confidence interval to a preset fuzzy inference model for fuzzy inference, wherein, through dynamically adjusting the weights of the fuzzy rules and the judgment confidence, a hierarchical warning result of the high-speed rail rail insulation joint is obtained.

[0102] It should be noted that for the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0103] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0104] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A high-speed rail insulation joint health monitoring and early warning method, characterized in that: include: Obtaining resistance data and environmental influencing factors of the rail insulation joint, wherein the environmental influencing factors include temperature data, humidity data and vibration data of the rail insulation joint; Based on the random projection algorithm, the resistance data of the rail insulation joint and the environmental influencing factors are reduced in dimension to obtain the resistance characteristic data after dimension reduction. Anomaly detection is performed on the resistance feature data after dimension reduction based on the isolation forest algorithm, wherein the labeled data containing resistance anomaly points are obtained through weight adjustment and dynamic optimization of path length; Based on the empirical mode decomposition method, the marked data containing the resistance abnormal point is decomposed, wherein the multi-scale intrinsic mode function feature data is obtained by constrained decomposition and instantaneous feature extraction of the marked data containing the resistance abnormal point; The multi-scale intrinsic mode function characteristic data is sent to the preset Gaussian process regression model for trend prediction, wherein the predicted value of resistance change and the corresponding confidence interval are obtained through kernel function combination and hyperparameter adjustment; The predicted value of resistance change and the corresponding confidence interval are sent to the preset fuzzy reasoning model for fuzzy reasoning, wherein the graded warning results of high-speed rail insulation joints are obtained by dynamically adjusting the weights of fuzzy rules and the confidence of judgment.

2. The high-speed rail insulation joint health monitoring and early warning method according to claim 1 is characterized in that ,The dimensionality reduction processing of the resistance data and environmental influencing factors of the rail insulation joint based on the random projection algorithm is performed to obtain the resistance characteristic data after dimensionality reduction, including: The resistance data and environmental influencing factors of the rail insulation joint are standardized, wherein the resistance data and environmental influencing factors of the rail insulation joint are standardized by a Z-score standardization method to obtain standardized high-dimensional resistance and environmental factor joint characteristic data; Generate a random Gaussian matrix based on the above joint feature data, wherein a sparse matrix is ​​randomly generated through Gaussian distribution to map the feature vector of the original high-dimensional space to obtain a random mapping matrix; Based on the random projection algorithm, the random mapping matrix is ​​processed by feature mapping, and the high-dimensional feature vector is mapped to the low-dimensional space to obtain the resistance feature data after preliminary dimensionality reduction. The resistance characteristic data after the preliminary dimensionality reduction is subjected to feature importance reconstruction processing, and the resistance characteristic data after the preliminary dimensionality reduction is optimized and adjusted by a weighted principal component reconstruction algorithm to obtain the final resistance characteristic data after dimensionality reduction.

3. The high-speed rail insulation joint health monitoring and early warning method according to claim 1 is characterized in that The method of performing anomaly detection on the resistance characteristic data after dimension reduction based on the isolation forest algorithm includes: Initializing the reduced-dimensional resistance characteristic data, wherein a basic isolation forest model is obtained by constructing a decision tree set based on random samples and defining an abnormality scoring formula; The basic isolation forest model is weighted, wherein the contribution of the resistance feature dimension is calculated and dynamically weighted to obtain a weighted isolation forest model; The weighted isolation forest model is subjected to a dynamic optimization process of path length, wherein an optimized anomaly scoring model is obtained by adjusting the splitting depth and the node coverage area; Based on the optimized abnormal scoring model, all the resistance feature data after dimension reduction are scored and thresholded to obtain marked data containing resistance abnormal points.

4. The high-speed rail insulation joint health monitoring and early warning method according to claim 1 is characterized in that , based on the empirical mode decomposition method, the labeled data containing the resistance abnormal point is decomposed, wherein the labeled data containing the resistance abnormal point is decomposed with constraints and the instantaneous feature is extracted, including: The marked data containing the abnormal resistance point is subjected to initial empirical mode decomposition processing, wherein a constrained decomposition strategy for extracting the intrinsic mode function is used to obtain a set of initially decomposed intrinsic mode functions; The initially decomposed intrinsic mode function set is subjected to noise suppression processing, wherein the intrinsic mode function components with low signal-to-noise ratio are removed by a singular value decomposition method to obtain a denoised intrinsic mode function set; The instantaneous frequency and amplitude of all intrinsic mode function components are analyzed by Hilbert transform, and the instantaneous characteristic matrix is ​​constructed; The instantaneous features in the instantaneous feature matrix are decomposed and reconstructed by wavelet packet transform to obtain multi-scale intrinsic mode function feature data.

5. The high-speed rail insulation joint health monitoring and early warning method according to claim 1 is characterized in that ,The multi-scale intrinsic mode function feature data is sent to the preset Gaussian process regression model for trend prediction, including: Performing kernel mapping processing on the multi-scale intrinsic mode function feature data, wherein high-dimensional feature data after kernel mapping is obtained by high-dimensional feature space mapping based on kernel function combination; The high-dimensional feature data after the kernel mapping is sent to a preset Gaussian process regression model for training, wherein the kernel matrix of the high-dimensional feature data after the kernel mapping is calculated and the hyperparameters are optimized to obtain an optimized Gaussian process model; The optimized Gaussian process model is used to predict the trend of resistance change, wherein the predicted value of the resistance change trend is obtained by calculating the posterior distribution; The confidence interval of the predicted value of the resistance change trend is dynamically optimized to obtain a predicted result with an abnormal mark and a corresponding confidence interval.

6. A high-speed rail insulation joint health monitoring and early warning system, characterized in that: include: An acquisition unit, used to acquire resistance data and environmental influencing factors of the rail insulation joint, wherein the environmental influencing factors include temperature data, humidity data and vibration data of the rail insulation joint; A processing unit, used for performing dimension reduction processing on the resistance data of the rail insulation joint and the environmental influencing factors based on a random projection algorithm to obtain resistance characteristic data after dimension reduction; A detection unit, used for performing anomaly detection on the resistance characteristic data after dimension reduction based on an isolation forest algorithm, wherein the marked data containing the resistance abnormal point is obtained through weight adjustment and dynamic optimization of the path length; An analysis unit, used for decomposing the marked data containing the abnormal resistance point based on the empirical mode decomposition method, wherein multi-scale intrinsic mode function feature data is obtained by performing constrained decomposition and instantaneous feature extraction on the marked data containing the abnormal resistance point; A prediction unit is used to send multi-scale intrinsic mode function feature data to a preset Gaussian process regression model for trend prediction, wherein the predicted value of resistance change and the corresponding confidence interval are obtained through kernel function combination and hyperparameter adjustment; The judgment unit is used to send the predicted value of the resistance change and the corresponding confidence interval to a preset fuzzy reasoning model for fuzzy reasoning, wherein the graded warning results of the high-speed railway rail insulation joint are obtained by dynamically adjusting the weights and judgment confidence of the fuzzy rules.

7. The high-speed railway rail insulation joint health monitoring and early warning system according to claim 6 is characterized in that: The processing unit comprises: A first processing subunit is used to perform standardization processing on the resistance data and environmental influencing factors of the rail insulation joint, wherein the resistance data and environmental influencing factors of the rail insulation joint are standardized by a Z-score standardization method to obtain standardized high-dimensional resistance and environmental factor joint characteristic data; A second processing subunit is used to generate a random Gaussian matrix based on the joint feature data, wherein a sparse matrix is ​​randomly generated by Gaussian distribution to map the feature vector of the original high-dimensional space to obtain a random mapping matrix; The third processing subunit is used to perform feature mapping processing on the random mapping matrix based on a random projection algorithm, map the high-dimensional feature vector to a low-dimensional space, and obtain resistance feature data after preliminary dimensionality reduction; The fourth processing subunit is used to perform feature importance reconstruction processing on the resistance characteristic data after the preliminary dimensionality reduction, and optimize and adjust the resistance characteristic data after the preliminary dimensionality reduction through a weighted principal component reconstruction algorithm to obtain the final resistance characteristic data after dimensionality reduction.

8. The high-speed railway rail insulation joint health monitoring and early warning system according to claim 6 is characterized in that: The detection unit comprises: A first detection subunit is used to initialize the resistance characteristic data after dimension reduction, wherein a basic isolation forest model is obtained by constructing a decision tree set based on random samples and defining an abnormality scoring formula; The second detection subunit is used to adjust the weight of the basic isolation forest model, wherein the weighted isolation forest model is obtained by calculating the contribution of the resistance feature dimension and dynamically allocating the weight; A third detection subunit is used to perform path length dynamic optimization processing on the weighted isolation forest model, wherein an optimized anomaly scoring model is obtained by adjusting the splitting depth and the node coverage area; The fourth detection subunit is used to score and divide the thresholds of all the resistance feature data after dimensionality reduction based on the optimized abnormality scoring model to obtain marked data containing resistance abnormal points.

9. The high-speed railway rail insulation joint health monitoring and early warning system according to claim 6, characterized in that: The analysis unit comprises: The first analysis subunit is used to perform initial empirical mode decomposition processing on the marked data containing the abnormal resistance point, wherein a constrained decomposition strategy of extracting the intrinsic mode function is used to obtain a set of initially decomposed intrinsic mode functions; The second analysis subunit is used to perform noise suppression processing on the initially decomposed intrinsic mode function set, wherein the intrinsic mode function components with low signal-to-noise ratio are removed by a singular value decomposition method to obtain a denoised intrinsic mode function set; The third analysis subunit is used to analyze the instantaneous frequency and amplitude of all intrinsic mode function components through Hilbert transform, and construct an instantaneous characteristic matrix; The fourth analysis subunit is used to perform frequency band decomposition and reconstruction on the instantaneous features in the instantaneous feature matrix by wavelet packet transform to obtain multi-scale intrinsic mode function feature data.

10. The high-speed railway rail insulation joint health monitoring and early warning system according to claim 6, characterized in that: The prediction unit comprises: The first prediction subunit is used to perform kernel mapping processing on the multi-scale intrinsic mode function feature data, wherein high-dimensional feature data after kernel mapping is obtained by high-dimensional feature space mapping based on kernel function combination; A second prediction subunit is used to send the high-dimensional feature data after the kernel mapping to a preset Gaussian process regression model for training, wherein the kernel matrix of the high-dimensional feature data after the kernel mapping is calculated and the hyperparameters are optimized to obtain an optimized Gaussian process model; A third prediction subunit is used to predict the trend of resistance change using the optimized Gaussian process model, wherein the predicted value of the resistance change trend is obtained by calculating the posterior distribution; The fourth prediction subunit is used to dynamically optimize the confidence interval of the predicted value of the resistance change trend to obtain a prediction result with an abnormal mark and a corresponding confidence interval.

Images