A railway turnout state early warning method, system, device and medium

Abstract

This invention belongs to the field of rail transit technology and provides a method, system, electronic device, and storage medium for early warning of railway turnout status. The method includes: preprocessing historical power curve data of the degraded state and the fault state of the turnout switch machine; performing cluster analysis on the preprocessed historical power curve data of the degraded state to obtain a degraded state classification; constructing a degraded state-fault relationship model based on the similarity between representative historical power curve data of the degraded state classification and the preprocessed historical power curve data of the fault state; predicting and classifying the power curve data of the degraded state of the turnout switch machine according to a neural network model to obtain a predicted degraded state classification; and using the degraded state-fault relationship model, assessing and issuing early warnings about the degraded state of the turnout switch machine based on the predicted degraded state classification. This makes railway transportation safer and more reliable.

Description

Technical Field

[0001] This invention belongs to the field of rail transit technology, and in particular relates to a method, system, electronic device and storage medium for early warning of railway turnout status. Background Technology

[0002] To assess the operational status of railway turnouts, a clear metric needs to be established. The turnout switch machine power curve is one of the important indicators reflecting the working status of the turnout, and it can intuitively show the operation process of the turnout switch machine. Under normal circumstances, the operation process of the turnout switch machine can be divided into three stages: unlocking, switching, and locking. When a fault occurs, the three stages of the power curve will show corresponding abnormal changes depending on the type of fault.

[0003] The degradation state of railway turnouts refers to the process by which the performance of turnout equipment gradually declines during long-term use due to wear, aging, environmental factors, etc., but has not yet reached a fault state. The fault state, on the other hand, refers to the state where the turnout equipment can no longer function properly, affecting the safety of railway operation. The degradation state and the fault state are closely related; the degradation state is a precursor to a fault, while the fault state is the result of the continued development of the degradation state. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a railway turnout status early warning method, system, electronic equipment, and storage medium. By deeply studying the relationship between turnout degradation and fault states, and calculating the similarity between different states, the operating status of the equipment is assessed. This enables real-time assessment of the turnout's operating status, allowing for timely measures to be taken for adverse conditions to ensure the safety and stability of railway transportation. Furthermore, by exploring the potential connections between turnout degradation states, predicting and assessing future operating states, and achieving proactive early warning, the accuracy and reliability of railway turnout early warning systems are improved.

[0005] To address the aforementioned technical problems, the first aspect of this invention proposes a railway turnout status early warning method, the method comprising: Preprocessing is performed on the historical power curve data of the turnout switch machine in both its degraded and fault states. Cluster analysis is performed on the preprocessed historical power curve data of the degradation state to obtain the degradation state classification; Based on the similarity between representative historical power curve data of the degradation state classification and the preprocessed historical power curve data of the fault state, a model of the relationship between degradation state and fault is constructed. Based on the neural network model, the power curve data of the degradation state of the turnout switch machine is predicted and classified, and the predicted degradation state classification is obtained. Using the degradation state and fault relationship model, the degradation state of the turnout switch machine is assessed and warned based on the predicted degradation state classification.

[0006] According to a preferred embodiment of the present invention, the preprocessing of historical power curve data of the degraded state and the fault state of the turnout switch machine includes: For the historical power curve data of the aforementioned degradation state, it is necessary to collect historical power curve data of multiple switch machines of the same model within the same time period; The historical power curve data of the fault state is set to correspond to a preset fault type.

[0007] According to a preferred embodiment of the present invention, the similarity between the representative historical power curve data based on the degradation state classification and the preprocessed historical power curve data of the fault state is used to construct a degradation state-fault relationship model, including: Based on the similarity, the classified degradation states are categorized into several working state types; Based on the degradation state after classification according to the relational model, the corresponding working state type and the maintenance suggestion corresponding to the working state type are obtained.

[0008] According to a preferred embodiment of the present invention, the step of predicting the power curve data of the degraded state of the turnout switch machine based on a neural network model includes: The power curve data of the degraded state is processed using the sliding window method and divided into several data sequences, including the training set; By using a neural network model to explore the potential relationships between data in the training set, the power curve data of the degradation state in the next working cycle is predicted.

[0009] According to a preferred embodiment of the present invention, the data sequence includes a training set, comprising: The data sequence is divided into a sample set, which also includes a validation set and a test set; The optimal hyperparameters of the neural network model are adjusted and determined using the validation set. The performance of the neural network model after the hyperparameters were determined was evaluated using the test set.

[0010] According to a preferred embodiment of the present invention, the degradation state of a turnout switch machine is assessed and warned based on the degradation state classification using the degradation state-fault relationship model, including: The degradation state classification with the highest similarity to the predicted degradation state classification is obtained, and the corresponding working state type is used as the predicted turnout working state. The fault type with the highest similarity to the predicted degradation state is identified, and a fault warning is issued.

[0011] To address the aforementioned technical problems, a second aspect of this invention proposes a railway turnout status early warning system, the system comprising: a data processing module, a data analysis module, a model building module, a fault prediction module, and a fault early warning module; The data processing module is used to preprocess the historical power curve data of the degraded state and the historical power curve data of the fault state of the turnout switch machine. The data analysis module is used to perform cluster analysis on the preprocessed historical power curve data of the degradation state to obtain the degradation state classification. The model building module is used to construct a model of the relationship between degradation state and fault based on the similarity between representative historical power curve data of degradation state classification and preprocessed historical power curve data of the fault state. The fault prediction module is used to predict and classify the power curve data of the degraded state of the turnout switch machine according to the neural network model, and obtain the predicted degraded state classification. The fault early warning module is used to assess and warn of the degradation status of the turnout switch machine by utilizing the degradation status and fault relationship model and classifying the predicted degradation status.

[0012] According to a preferred embodiment of the present invention, it includes: The data processing module is also used to collect historical power curve data of multiple switch machines of the same model within the same time period for the historical power curve data of the degradation state. The data processing module is also used to set the historical power curve data of the fault state to correspond to a preset fault type.

[0013] According to a preferred embodiment of the present invention, it includes: The model building module is also used to classify the degenerate state after classification into several working state types based on the similarity. The model building module is also used to obtain the corresponding working state type and the maintenance suggestion corresponding to the working state type based on the degradation state after the input classification of the relation model.

[0014] According to a preferred embodiment of the present invention, it includes: The fault prediction module is also used to process the power curve data of the degradation state using the sliding window method, and divide it into several data sequences, the data sequences including a training set; The fault prediction module is also used to explore the potential relationships between the data in the training set using a neural network model, and to predict the power curve data of the degradation state in the next working cycle.

[0015] According to a preferred embodiment of the present invention, it includes: The fault prediction module is also used to divide the data sequence into sample sets, including a validation set and a test set; The fault prediction module is also used to adjust and determine the optimal hyperparameters of the neural network model using the validation set; The fault prediction module is also used to evaluate the performance of the neural network model after the hyperparameters have been determined using the test set.

[0016] According to a preferred embodiment of the present invention, it includes: The fault warning module is also used to obtain the degradation state classification with the highest similarity to the predicted degradation state classification, and the corresponding working state type is used as the predicted turnout working state. The fault warning module is also used to determine the fault type with the highest similarity to the predicted degradation state classification, and to issue a fault warning.

[0017] To address the aforementioned technical problems, a third aspect of the present invention provides an electronic device, comprising: processor; And a memory storing computer-executable instructions, which, when executed, cause the processor to perform the method described in any of the above embodiments.

[0018] To address the aforementioned technical problems, a fourth aspect of the present invention provides a computer storage medium, wherein the computer storage medium stores one or more programs, which, when executed by a processor, implement the method described in any of the above embodiments.

[0019] Compared with existing technologies, this invention has the following advantages: It utilizes clustering analysis to classify the degradation states of turnouts, clarifying the characteristics of different degradation states. It calculates the similarity between degradation states and fault states, providing a basis for evaluating the turnout's operational status. It employs deep learning algorithms to identify and comprehensively evaluate the turnout's operational status, achieving real-time monitoring of turnout degradation states. It uses the Transformer model in deep learning to predict the turnout's status in the next working cycle, thereby achieving early prevention and effective control of faults. This forms a complete method for turnout status evaluation and early warning, providing technical support for the safety and reliability of railway transportation.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown; Figure 2 A second schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown. Figure 3 A schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown in part three. Figure 4 A schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown in Figure 4. Figure 5 A schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown in Figure 5. Figure 6 A schematic flowchart of a railway turnout status early warning method according to an embodiment of the present invention is shown in Figure 6. Figure 7 A structural diagram of a railway turnout status early warning system according to an embodiment of the present invention is shown; Figure 8 A schematic diagram of an electronic device structure according to an embodiment of the present invention is shown; Figure 9 The power curve of the turnout switch machine is shown; Figure 10 The power curve corresponding to the turnout switching jamming fault is shown; Figure 11 The flowchart of cluster analysis is shown; Figure 12 A flowchart illustrating the construction process of the model relating degradation state to fault is shown. Figure 13 A schematic diagram of a sliding window is shown; Figure 14 A schematic diagram of the Attention mechanism is shown. Figure 15 A diagram illustrating the working principle of the Self-Attention mechanism is shown. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The same reference numerals in the accompanying drawings denote the same or similar elements, components, or parts, and therefore, repeated descriptions of the same or similar elements, components, or parts may be omitted below. It should also be understood that although terms such as first, second, third, etc., indicating numbers may be used herein to describe various devices, elements, components, or parts, these devices, elements, components, or parts should not be limited by these terms. That is, these terms are only used to distinguish one from another. For example, a first device may also be referred to as a second device, without departing from the essential technical solution of the invention. Furthermore, the terms "and / or" and "and / or" refer to all combinations including any one or more of the listed items.

[0025] like Figure 1 As shown, the method includes: S11. Preprocess the historical power curve data of the turnout switch machine in the degraded state and the historical power curve data in the fault state.

[0026] like Figure 9 As shown in this embodiment, under normal circumstances, the operation of the turnout switch machine can be divided into three stages: unlocking, switching, and locking. Each stage has a corresponding performance on the power curve. For example, the power curve will show a steep peak during the unlocking stage, the power curve during the switching stage will be relatively flat, and the power curve will drop sharply to zero during the locking stage.

[0027] like Figure 10 As shown in this embodiment, when a fault occurs, the three stages of the power curve will exhibit corresponding abnormal changes depending on the fault type. For example, Figure 2 The diagram shows the power curve corresponding to a turnout switching jamming fault. The power curve shows that the turnout functions normally in the unlocked state, but experiences significant fluctuations during the switching state, followed by a sudden increase and then stabilization. Simultaneously, when a turnout switching jamming fault occurs, the turnout switch machine's operating time is extended compared to normal conditions, which is also reflected in the power curve.

[0028] In this embodiment, the degradation state of a railway turnout refers to the process by which the performance of the turnout equipment gradually declines during long-term use due to wear, aging, environmental factors, etc., but has not yet reached a fault state. The fault state, on the other hand, refers to the state where the turnout equipment can no longer function properly, affecting railway operation safety. The degradation state and the fault state are closely related; the degradation state is a precursor to a fault, while the fault state is the result of the continued development of the degradation state.

[0029] In this embodiment, a Savitzky-Golay filter is used to denoise the power curve data to reduce the impact of noise on data analysis. The Savitzky-Golay filter is an effective data smoothing technique that smooths the data by fitting a polynomial within a local window of the data points, while preserving the data's characteristics as much as possible.

[0030] In this embodiment, given that the power curve of the turnout switch machine is time-series data, the LLTB (Largest Triangle Three Buckets) algorithm is used for feature extraction. The LLTB algorithm is an effective downsampling tool for time-series data; it improves data processing efficiency while maintaining data integrity by intelligently selecting representative data points. Furthermore, LLTB can effectively capture the key shape and trend of the curve. Therefore, this invention uses the dimensionality reduction algorithm LLTB to directly extract features from the power curve of the turnout switch.

[0031] In this embodiment, to facilitate subsequent cluster analysis, similarity analysis, degradation state classification, and prediction, the extracted feature data needs to be normalized. This embodiment can use the Z-Score Normalization algorithm to normalize the extracted feature data.

[0032] In this embodiment, the data processing and feature extraction of this module are the foundation of the data used in the entire scheme. Subsequent clustering analysis, similarity analysis, degradation state identification and prediction modules will conduct further analysis and processing based on this.

[0033] S12. Perform cluster analysis on the preprocessed historical power curve data of the degradation state to obtain the degradation state classification.

[0034] In this embodiment, cluster analysis is performed on the data after preprocessing and feature extraction to classify the degradation status of the turnout switch machine. Since the turnout power curve is a time series, the k-Shape method can be used in this embodiment.

[0035] like Figure 11 As shown, in this embodiment, the steps of cluster analysis can be as follows: (1) Initialize data: Randomly divide all time series into k clusters. The cluster membership relationship remains unchanged or the maximum allowed number of iterations is reached.

[0036] (2) Calculate the centroids of k clusters using the SBD distance metric.

[0037] (3) Use the SBD distance metric to calculate the distance between each time series and the k cluster centers.

[0038] (4) Assign each time series to the nearest cluster center based on its distance from the cluster center.

[0039] (5) Repeat steps (2), (3), and (4) until the cluster membership relationship remains unchanged or the maximum allowed number of iterations is reached.

[0040] In this embodiment, the k-Shape algorithm, based on the K-means clustering concept, innovates on distance metric and centroid calculation, thus considering the shape of the time series. k-Shape relies on a scalable iterative refinement process that creates uniform and well-separated clusters. As a distance metric, k-Shape uses a normalized version of the cross-correlation metric to consider the shape of the time series while comparing them. Based on the properties of the distance metric, a method for calculating cluster centroids is proposed, which is used in each iteration to update the cluster assignments of the time series.

[0041] In this embodiment, k-Shape uses a normalized version of the cross-correlation distance metric, SBD, to compare time series. This method considers the shape of the time series and is invariant to scaling and shifting. In the k-Shape clustering algorithm, SBD is used for two aspects: assigning time series to clusters and updating cluster centers. Assigning time series to clusters: By calculating the SBD between the time series and the cluster center, the time series is assigned to the nearest cluster. Updating cluster centers: During the iteration process, SBD is used to determine new cluster centers to minimize the sum of the SBDs of all time series within a cluster with respect to the cluster center.

[0042] In this embodiment, the iterative refinement process involves the algorithm continuously updating the centroids of the clusters through an iterative process until convergence or the maximum number of iterations is reached. Cluster centroid calculation: k-Shape employs a method based on cross-correlation metrics to calculate cluster centroids, which are used in each iteration to update the time series' assignment to the clusters. Efficiency: The k-Shape algorithm scales linearly with the number of time series and generates homogeneous and well-separated classifications, making it highly efficient for handling large-scale time series datasets.

[0043] In this embodiment, before iteration, k time series are randomly selected as initial cluster centers. Then, through iteration, all time series are divided into k clusters, or k categories. During continuous iteration, the k-Shape algorithm estimates the distance between all time series and the cluster centers using the SBD algorithm, assigning the time series to the nearest cluster. Then, k-Shape updates the cluster centers so that the center's shape is as close as possible to the average shape of all time series within the cluster. This process iterates until convergence or the maximum number of iterations is reached.

[0044] In this embodiment, the advantage of k-Shape in time series clustering is mainly reflected in its efficient and accurate time series distance metric algorithm. The distance metric algorithm between two time series can also be applied to similarity analysis and classification. Therefore, in this invention, the SBD metric algorithm in the k-Shape algorithm is applied not only to similarity analysis but also to degradation state identification.

[0045] In this embodiment, predicting the power curve of railway turnouts can help maintenance personnel detect potential faults in turnout equipment in advance, thereby taking corresponding maintenance measures to avoid the impact of faults on railway transportation. Although there are currently few methods for predicting railway turnout curves, methods for predicting equipment power curves in other fields have been studied to some extent, such as power systems and mechanical equipment. These studies mostly use deep learning algorithms such as LSTM and GRU to learn from historical data, find potential correlations between data, and thus predict future power curves. Therefore, this invention will draw on these methods to predict the power curve of railway turnouts, achieving early warning of the turnout's operating status. In summary, predicting the power curve of the turnout switch machine in the next working cycle is feasible and meaningful. By predicting the equipment power curve, abnormal conditions of the equipment can be detected in advance, enabling preventive maintenance and improving the reliability and safety of the equipment.

[0046] In this embodiment, in reality, the turnout only operates when a train passes by, and the intervals between trains passing the turnout are not fixed. Therefore, the collected data may exhibit temporal irregularities. In other words, if the power curve of one turnout operation is considered as a data point, then the time intervals between multiple data points are not fixed. In this case, traditional time series forecasting methods may not be applicable because these methods typically assume that the time intervals between data points are fixed.

[0047] S13. Based on the similarity between representative historical power curve data of the degradation state classification and the preprocessed historical power curve data of the fault state, construct a relationship model between degradation state and fault.

[0048] In this embodiment, the degradation status assessment and early warning module mainly uses the constructed degradation status and fault relationship model to assess and warn of the current turnout working status. The input of the module is the degradation status to which the current status belongs, and the output is the maintenance suggestions and corresponding fault types given for the current working status.

[0049] S14. Based on the neural network model, predict the power curve data of the degradation state of the turnout switch machine, and perform classification analysis to obtain the predicted degradation state classification.

[0050] In this embodiment, the module mainly predicts and classifies the power curve of the turnout equipment in the degraded state to obtain the predicted degraded state classification.

[0051] S15. Using the degradation state and fault relationship model, the degradation state of the turnout switch machine is assessed and warned based on the predicted degradation state classification.

[0052] In this embodiment, the current working state of the turnout is evaluated and warned by the constructed degradation state and fault relationship model. The input of the module is the predicted degradation state classification, and the output is the maintenance suggestions and corresponding fault types given for the predicted working state.

[0053] like Figure 2 As shown, the method includes: S21. For the historical power curve data of the degradation state, it is necessary to collect the historical power curve data of multiple switch machines of the same model within the same time period.

[0054] In this embodiment, the methods for collecting power curve data of turnout switch machines generally include three types: (1) Using a current power sensor: This method calculates the power value by collecting the voltage and current values ​​when the motor is running. The specific steps are as follows: Three-phase voltages are collected at terminals 11, 31, and 51 of the front-end of the phase loss protector (DBQ).

[0055] The three-phase current is collected between the DBQ output and the 1DQJ (1DQJF) contact.

[0056] When 1DQJ is activated, the current transformer is started to collect the voltage and current values ​​of the motor during operation. The active power is calculated every 40 milliseconds and recorded until the operation ends.

[0057] (2) Use a dedicated three-phase power chip: For example, use the ATT7022 chip, combined with a high-speed, pipelined MCU system, to form the core of the power sensor device. This method is suitable for harsh working environments with high interference.

[0058] (3) Using an embedded processor: For example, using STM32F103xx as the main control chip to design a switch machine power acquisition unit. This acquisition unit can measure voltage, current and power, and transmit data through the RS-485 communication interface of the Modbus RTU protocol, and directly network with terminals such as PLC and RTU.

[0059] S22. Set the historical power curve data of the fault state to correspond to the preset fault type.

[0060] In this embodiment, the diagnosis of turnout faults is not involved, so it is assumed that the fault type corresponding to the power curve under fault conditions is known.

[0061] like Figure 3 As shown, the method includes: S31. Based on the similarity, the classified degradation states are categorized into several working state types.

[0062] In this embodiment, as Figure 12 As shown, the process of constructing the model of the relationship between degradation state and fault is as follows: (1) Obtain the k-Shape algorithm after iterative refinement Cluster centers, and these cluster centers are used as A representative of a degenerate state, assuming The cluster center is Cluster center For a time series (power curve) .

[0063] (2) Assume that Types of fault states The power curve for the l-th fault state is as follows: ,in This is a time series (power curve) \(q=1,...,n\). This data has been processed by the data preprocessing and feature extraction modules.

[0064] (3) Calculate the similarity between each pair of degraded states and fault states. Since the amount of data for each fault state is small, this invention will cluster the data. The representative power curve for each degradation state and all power curves for each fault state are used to calculate the similarity between the degradation state and the fault state. The average of the obtained SBD values ​​is then used as the similarity between the two states. The calculation process is as follows: ① Degenerative state and There are between the various fault states For each combination, calculate its similarity.

[0065] Specifically, the first The first degenerative state and the second The similarity between the fault states is , , : ; in, For the first The first degenerative state and the second The average SBD value between the various fault states Indicates the first Representative power curves for each degradation state. Indicates the first The first fault state A power curve, Indicates the first The representative power curve of the first degradation state and the second The first fault state SBD values ​​between the power curves.

[0066] ②Because the range of values ​​for SBD is Therefore, the range of values ​​for the similarity SM is also... The smaller the SM value, the more similar the two power curves are.

[0067] (4) The degradation states are summarized according to the range of SBD values. In this embodiment, the turnout working states are divided into four types according to the range of SBD values, which are called "working state types". The working states corresponding to the turnout are fault ([0,0.5]), caution ([0.5,1]), good ([1,1.5]), and healthy ([1.5,2]). The k degradation states are summarized according to the interval to which the calculated similarity SM belongs. At the same time, the fault type most similar to the current degradation state is given as a warning identification of fault type.

[0068] (5) The maintenance corresponding to the four working status types can be as follows: Fault: The turnout exhibits obvious fault characteristics and requires immediate repair; Note: Turnout faults may not have obvious characteristics and require inspection and close monitoring. Good: The turnout status parameters deviate from the rated values, but are still within the normal range, and the probability of failure is controllable; Health: The turnout status parameters are normal, no maintenance is required, and the probability of failure is low.

[0069] In this embodiment, the SBD (Shape-based distance) metric algorithm is a method within the k-Shape algorithm used to calculate time series similarity. The SBD algorithm is specifically designed to process time series data; it captures the shape characteristics of the time series through normalized cross-correlation, rather than simply relying on numerical differences. In this invention, the SBD algorithm is used to analyze the similarity between the switch machine power curves under degraded and faulty states, obtaining the corresponding SBD values. Then, the operating status of the turnout is evaluated based on the SBD values; a smaller SBD value indicates a stronger correlation between the two power curves.

[0070] In this embodiment, the SBD algorithm is based on the cross-correlation metric CC, a method for measuring the similarity between two time series, especially when they may have temporal offsets. The cross-correlation metric is calculated by sliding one series over time and comparing it to another to find the optimal alignment.

[0071] Specifically, for two time series of equal length (or unequal length) and ,Keep Stay still, slide sequentially with The inner product is calculated for the corresponding points, and the sliding window size is... : ; in .

[0072] Specifically, the cross-correlation sequence is defined as follows: , Its length is .

[0073] The specific calculation formula is as follows: , .in The calculations are as follows: ; The goal is to find The position makes Maximum. Based on this... , Compared to Optimal offset That will determine it, among which .

[0074] In this embodiment, to make the cross-correlation metric more robust and effective, the k-Shape algorithm is used. Normalization was performed. The normalization process involves... Value scaled to Within this range, this helps to eliminate amplitude differences between different time series and focus on shape comparisons.

[0075] Specifically, the obtained normalized cross-correlation measure is: .

[0076] In this embodiment, to improve computational efficiency, the SBD algorithm utilizes the Fast Fourier Transform (FFT) to calculate cross-correlation. By converting the time series to the frequency domain, the convolutions between them (a form of cross-correlation metric) can be calculated quickly, thereby significantly reducing computation time.

[0077] In this embodiment, the calculation of the SBD algorithm includes the following steps: (1) Preprocessing: Perform Z-Score Normalization on the time series to ensure that the mean of each series is 0 and the standard deviation is 1.

[0078] (2) Calculate the cross-correlation measure: For each pair of time series, calculate their cross-correlation sequence.

[0079] (3) Normalization: Divide the cross-correlation sequence by the geometric mean of the autocorrelation of the two sequences (i.e., the cross-correlation measure of the sequence with itself) to obtain the normalized cross-correlation measure.

[0080] (4) Distance metric: It is defined by subtracting the maximum value of the normalized cross-correlation measure from 1. The smaller this value, the more similar the two time series are.

[0081] Specifically, The calculation formula is as follows: ; in, The range is Between these two values, the smaller the value, the more similar the two sequences are. This indicates that the time series are completely similar.

[0082] In this embodiment, the advantage of the SBD algorithm lies in its ability to accurately capture the shape characteristics of time series, unaffected by changes in the amplitude, offset, or velocity of the series. This makes it well-suited for processing time series data with complex patterns and dynamic changes.

[0083] S32. Based on the degradation state after classification according to the relational model, obtain the corresponding working state type and the maintenance suggestion corresponding to the working state type.

[0084] In this embodiment, the relational model can obtain the corresponding working state type and the maintenance suggestions corresponding to the working state type based on the input degradation state type.

[0085] In this embodiment, the degradation state can be identified using newly acquired turnout power curve data. It is only necessary to estimate the SBD value between the current state and the known k degradation states, select the most similar (smallest SBD value) degradation state as the current turnout working state, and then obtain maintenance suggestions for the current turnout equipment based on the previously calculated similarity relationship between the degradation state and the fault state.

[0086] like Figure 4 As shown, the method includes: S41. The power curve data of the degraded state is processed using the sliding window method and divided into several data sequences, the data sequences including the training set.

[0087] like Figure 13 As shown, in this embodiment, the sliding window method is used to process the power curve data of the turnout equipment. The long-term power curve data is divided into several shorter data sequences using the training set, which also increases the data volume. Assume that the turnout power curve of a certain working cycle is a sample point, the original data sequence length is L, the sliding window size is K, and the sliding step size is d. Through the sliding window mechanism, L is divided into multiple short sequences of the same length K. Since the original sequences have a temporal relationship, the short sequences segmented by the sliding window method also maintain the original temporal relationship. This invention inputs the first K samples into the model, and the model outputs the (K+1)th sample, which is the predicted power curve for the next working cycle.

[0088] S42. Utilize a neural network model to explore the potential relationships between the data in the training set and predict the power curve data of the degradation state for the next working cycle.

[0089] In this embodiment, the turnout power curve data used in this invention has a high dimensionality. Therefore, to simplify the internal structure of the model and improve training efficiency, this invention employs a Transformer model with two layers (Encoder-layer and Decoder-layer) to predict the power curve of the turnout equipment. The data processing procedure within the Transformer model will be described in detail below. The internal structure of the Transformer model is shown in the figure. The model includes an encoder and a decoder; the encoder's output needs to be input to each decoding layer in the decoder.

[0090] In this embodiment, for the encoder, the data first passes through an embedding layer to convert the input power curve data into a vector representation of the same dimension. Then, positional encoding is used to add temporal information to the data. Next, two encoding layers are used to explore potential relationships between the data, with a multi-head attention layer at its core. The encoder's output needs to be fed into each multi-head attention layer in the decoder, which then uses the encoder's output to predict the power curve for the next working cycle. In this embodiment, for the decoder, the input data includes the encoder's output and the predicted value from the previous time step. First, an embedding layer is used to convert the predicted value from the previous time step into a vector representation of the same dimension. Then, positional encoding is used to add temporal information to the data. Next, two encoding layers are used to predict the power curve for the next working cycle; in addition to the multi-head attention layers, the encoding layers also include a masked multi-head attention layer, which is mainly used to prevent the model from seeing future information during prediction. Finally, a linear mapping layer is used to map the decoder's output to the target time series.

[0091] In this embodiment, the Transformer algorithm is a deep learning model based on an attention mechanism. It was initially used in the field of natural language processing and has since been widely applied in fields such as time series prediction and image processing. The Transformer model is a new type of neural network architecture that differs from RNNs and CNNs. It abandons the recursive architecture of traditional RNNs and CNNs, relies entirely on the attention mechanism, and captures the relationships between data in one go.

[0092] In this embodiment, the self-attention mechanism in the Transformer enables the model to effectively capture long-term dependencies in time-series data, and this core mechanism is also well-suited for handling turnout power curve prediction. Long-term dependencies refer to the relationships between events or features that are geographically distant in the time-series data that the model needs to capture. Most turnouts are in normal operating condition, a state that this invention identifies as a degraded state. The transition between a degraded state and a fault state is a long-term process, and accurately predicting the degraded state of turnouts requires a model suitable for handling long-term dependencies. In addition, the Transformer's advantages lie in position encoding, parallel processing of sequence data, and its encoder-decoder structure, all of which give the Transformer model a certain advantage in processing turnout power curve data.

[0093] like Figure 14As shown in this embodiment, the Attention mechanism, as the name suggests, is an attention mechanism, similar to the attention allocation mechanism when a person observes something. It focuses limited attention on the most interesting (or most valuable) parts, thus saving resources and achieving the goal of quickly acquiring effective information. The Attention mechanism essentially obtains the coefficients of attention allocation, which can be simply summarized as "weighted summation." The overall principle of the Attention mechanism is to calculate the final result of the mapping relationship between the query and the key-value pair, i.e., the attention allocation coefficients.

[0094] In this embodiment, the self-attention mechanism proposed in the Transformer model differs from the general single-layer attention mechanism. In the self-attention mechanism, the Query, Key, and Value all come from the same set of input sequences. The working principle of the self-attention mechanism will be described in detail below. In neural networks, the self-attention mechanism can appear as a layer of the model.

[0095] like Figure 15 As shown, in this embodiment, the working process of the Self-Attention layer can be roughly divided into three stages. The first stage is to process the input sequence of the layer... With different weights , and Multiply to obtain the corresponding , and The second stage, as shown in formula (2.3), firstly calculates the similarity using the Similarity() function. and The similarity between them is then normalized using the SoftMax() function.

[0096] In this embodiment, the attention allocation coefficients obtained through the above process represent the elements in the input sequence. The relationships between elements are both overall and partial. The overall relationship is reflected in the first step of determining similarity. Similarity was calculated with all remaining elements in the sequence, and local connections were represented by attention allocation coefficients applied to the current element. Yes. This is also one of the advantages of the Attention mechanism, namely, it can obtain the overall and local connections at once. Below, we will briefly introduce some advantages of the Attention mechanism.

[0097] In this embodiment, the advantages of the Attention mechanism are obvious. First, compared to traditional RNN and CNN networks, the Attention mechanism has lower complexity and fewer parameters. Second, the calculation result of each step in the Attention mechanism does not affect subsequent calculations. Therefore, both the Attention mechanism and CNN networks have the ability to process tasks in parallel, thus compensating for the limitation of RNNs in parallel computation. However, CNN models only capture local information each time, and can only obtain the overall relationship by simply stacking layers. The Attention mechanism solves this problem well. It can flexibly capture the relationship between the whole and the parts, and this capture is completed in one step. In addition, the Attention mechanism is better than the RNN model for dealing with dependencies between elements. This is because, in the calculation process of the Attention mechanism, each element needs to be compared with other elements in the sequence. Therefore, the distance between each element in this process can be regarded as 1, which is much better than the RNN model, which needs to recursively obtain long-term dependencies step by step. Furthermore, as the length of the sequence increases, the long-term dependencies captured by the RNN model become weaker.

[0098] In this embodiment, the essence of the Transformer model lies in the Encoder-Decoder structure. The Encoder and Decoder of the Transformer model are each composed of six stacked Encoder-layers and Decoder-layers, and the overall output of the Encoder part needs to be input to each Decoder-layer in the Decoder.

[0099] In this embodiment, the Transformer model comprises two parts: an Encoder and a Decoder. Each Encoder-layer in the Encoder part includes two sub-layers: one is a Multi-head Attention layer, and the other is a fully connected simple feedforward neural network. After these two sub-layers, residual processing (i.e., the input and output of this layer are used together as the input to the next layer) is applied to the output, followed by Layer Normalization for normalization. Each Decoder-layer in the Decoder part includes three sub-layers, one more than the Encoder-layer (masked Multi-Head Attention layer). The remaining parts are similar to the Encoder-layer, and each sub-layer is also followed by residual connections and normalization.

[0100] In this embodiment, the core of the Transformer model lies in the previously mentioned Self-Attention and Multi-Head Attention mechanisms. The Multi-Head Attention mechanism in this model combines multiple different Self-Attention mechanisms. That is, it first processes multiple different sets of Q, K, and V through the Self-Attention mechanism, and then combines the processing results to obtain the multi-head attention allocation coefficients.

[0101] like Figure 5 As shown, the method includes: S51. The data sequence is divided into a sample set, which also includes a validation set and a test set.

[0102] In this embodiment, after processing the data using the sliding window method, the processed data file needs to be divided into a sample set to obtain a training set, a validation set, and a test set. By analyzing the data volume and characteristics, this embodiment can set the ratio of the training set, validation set, and test set to 8:1:1.

[0103] S52. Use the validation set to adjust and determine the optimal hyperparameters of the neural network model.

[0104] In this embodiment, the purpose of dividing the validation set is to adjust and determine the optimal model hyperparameters.

[0105] S53. Use the test set to evaluate the performance of the neural network model after the hyperparameters have been determined.

[0106] In this embodiment, the purpose of dividing the test set is to evaluate the performance of the final model and to approximate the generalization error of the model in the real case with the error obtained by the final model on the test set.

[0107] like Figure 6 As shown, the method includes: S61. Obtain the degradation state classification with the highest similarity to the predicted degradation state classification, and use the corresponding working state type as the predicted turnout working state.

[0108] In this embodiment, the working status type identification process is as follows: (1) Power curve data based on predicted degradation state classification and SBD values ​​between the power curve and the center of k clusters.

[0109] (2) Select the cluster with the smallest SBD value as the current working state of the turnout equipment.

[0110] S62. Determine the fault type with the highest similarity to the predicted degradation state classification, and issue a fault warning.

[0111] In this embodiment, the similarity between the predicted degradation state classification and the fault type set in S22 is calculated, the degradation state classification with the highest similarity is determined, and fault warning is given for the railway turnout status.

[0112] like Figure 7 As shown, a railway turnout status early warning system includes: a data processing module, a data analysis module, a model building module, a fault prediction module, and a fault early warning module.

[0113] In this embodiment, the data processing module is specifically used to preprocess the historical power curve data of the degraded state and the historical power curve data of the fault state of the turnout switch machine.

[0114] In this embodiment, the data analysis module is specifically used to perform cluster analysis on the preprocessed historical power curve data of the degradation state to obtain a degradation state classification.

[0115] In this embodiment, the model building module is specifically used to construct a model of the relationship between degradation state and fault by measuring the similarity between representative historical power curve data based on degradation state classification and preprocessed historical power curve data of the fault state.

[0116] In this embodiment, the fault prediction module is specifically used to predict and classify the power curve data of the degradation state of the turnout switch machine according to the neural network model, and obtain the predicted degradation state classification.

[0117] In this embodiment, the fault early warning module is specifically used to assess and warn of the degradation status of the turnout switch machine by utilizing the degradation status and fault relationship model and classifying the predicted degradation status.

[0118] In this embodiment, the data processing module is specifically used to collect historical power curve data of multiple switch machines of the same model within the same time period for the historical power curve data of the degraded state.

[0119] In this embodiment, the data processing module is specifically used to set the historical power curve data of the fault state to correspond to a preset fault type.

[0120] In this embodiment, the model building module is specifically used to classify the degraded states after classification into several working state types based on the similarity.

[0121] In this embodiment, the model building module is specifically used to obtain the corresponding working state type and the maintenance suggestion corresponding to the working state type based on the degradation state after inputting the relationship model.

[0122] In this embodiment, the fault prediction module is specifically used to process the power curve data of the degradation state using the sliding window method, dividing it into several data sequences, the data sequences including a training set.

[0123] In this embodiment, the fault prediction module is specifically used to explore the potential relationships between the data in the training set using a neural network model, and to predict the power curve data of the degradation state in the next working cycle.

[0124] In this embodiment, the fault prediction module is specifically used to divide the data sequence into sample sets, and also includes a validation set and a test set.

[0125] In this embodiment, the fault prediction module is specifically used to adjust and determine the optimal hyperparameters of the neural network model using the validation set.

[0126] In this embodiment, the fault prediction module is specifically used to evaluate the performance of the neural network model after the hyperparameters have been determined using the test set.

[0127] In this embodiment, the fault warning module is specifically used to obtain the degradation state classification with the highest similarity to the predicted degradation state classification, and the corresponding working state type is used as the predicted turnout working state.

[0128] In this embodiment, the fault warning module is specifically used to determine the fault type with the highest similarity to the predicted degradation state classification, and to issue a fault warning.

[0129] like Figure 8 As shown, this embodiment of the invention provides an electronic device, including a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. Memory 1130 is used to store computer programs; When the processor 1110 executes the program stored in the memory 1130, it implements any of the above-described determination methods.

[0130] The electronic device provided in this embodiment of the invention includes a processor 1110 that executes a program stored in a memory 1130 to obtain the fluid flow rate of each branch under different switching states and determines the initial volumetric flow rate of each branch; it corrects the initial volumetric flow rate based on the pipe parameters and fluid parameters of each branch when it is in operating condition and standard condition to obtain the standard condition volumetric flow rate of each branch; it obtains multiple total standard condition volumetric flow rates based on the standard condition volumetric flow rates of each branch under different switching states, and determines the optimal switching state of each branch by using the switching state of each branch when the total standard condition volumetric flow rate reaches a preset target.

[0131] The communication bus 1140 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, it is shown in the figure with only one thick line, but this does not indicate that there is only one bus or one type of bus.

[0132] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0133] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1130 may also be at least one storage device located remotely from the aforementioned processor 1110.

[0134] The processor 1110 mentioned above can be a general-purpose processor 1110, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0135] This invention provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors 1110 to implement the determination method of any of the above embodiments.

[0136] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0137] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of railway turnout status, characterized in that, The method includes: Preprocessing is performed on the historical power curve data of the turnout switch machine in both its degraded and fault states. Cluster analysis is performed on the preprocessed historical power curve data of the degradation state to obtain the degradation state classification; Based on the similarity between representative historical power curve data of the degradation state classification and the preprocessed historical power curve data of the fault state, a model of the relationship between degradation state and fault is constructed. Based on the neural network model, the power curve data of the degradation state of the turnout switch machine is predicted and classified, and the predicted degradation state classification is obtained. Using the degradation state and fault relationship model, the degradation state of the turnout switch machine is assessed and warned based on the predicted degradation state classification.

2. The early warning method according to claim 1, characterized in that, The preprocessing of historical power curve data for the degraded state and the fault state of the turnout switch machine includes: For the historical power curve data of the aforementioned degradation state, it is necessary to collect historical power curve data of multiple switch machines of the same model within the same time period; The historical power curve data of the fault state is set to correspond to a preset fault type.

3. The early warning method according to claim 1, characterized in that, The similarity between the representative historical power curve data based on the degradation state classification and the preprocessed historical power curve data of the fault state is used to construct a model of the relationship between degradation state and fault, including: Based on the similarity, the classified degradation states are categorized into several working state types; Based on the degradation state after classification according to the relational model, the corresponding working state type and the maintenance suggestion corresponding to the working state type are obtained.

4. The early warning method according to claim 1, characterized in that, The prediction of the power curve data of the degraded state of the turnout switch machine based on the neural network model includes: The power curve data of the degraded state is processed using the sliding window method and divided into several data sequences, including the training set; By using a neural network model to explore the potential relationships between data in the training set, the power curve data of the degradation state in the next working cycle is predicted.

5. The early warning method according to claim 4, characterized in that, The data sequence includes a training set, comprising: The data sequence is divided into a sample set, which also includes a validation set and a test set; The optimal hyperparameters of the neural network model are adjusted and determined using the validation set. The performance of the neural network model after the hyperparameters were determined was evaluated using the test set.

6. The early warning method according to claim 2 or 3, characterized in that, Using the degradation state and fault relationship model, and based on the predicted degradation state classification, the degradation state of the turnout switch machine is assessed and an early warning is issued, including: The degradation state classification with the highest similarity to the predicted degradation state classification is obtained, and the corresponding working state type is used as the predicted turnout working state. The fault type with the highest similarity to the predicted degradation state is identified, and a fault warning is issued.

7. A railway turnout status early warning system, characterized in that, The system includes: a data processing module, a data analysis module, a model building module, a fault prediction module, and a fault early warning module; The data processing module is used to preprocess the historical power curve data of the degraded state and the historical power curve data of the fault state of the turnout switch machine. The data analysis module is used to perform cluster analysis on the preprocessed historical power curve data of the degradation state to obtain the degradation state classification. The model building module is used to construct a model of the relationship between degradation state and fault based on the similarity between representative historical power curve data of degradation state classification and preprocessed historical power curve data of the fault state. The fault prediction module is used to predict and classify the power curve data of the degraded state of the turnout switch machine according to the neural network model, and obtain the predicted degraded state classification. The fault early warning module is used to assess and warn of the degradation status of the turnout switch machine by utilizing the degradation status and fault relationship model and classifying the predicted degradation status.

8. The early warning system according to claim 7, characterized in that, include: The data processing module is also used to collect historical power curve data of multiple switch machines of the same model within the same time period for the historical power curve data of the degradation state. The data processing module is also used to set the historical power curve data of the fault state to correspond to a preset fault type.

9. The early warning system according to claim 7, characterized in that, include: The model building module is also used to classify the degenerate state after classification into several working state types based on the similarity. The model building module is also used to obtain the corresponding working state type and the maintenance suggestion corresponding to the working state type based on the degradation state after the input classification of the relation model.

10. The early warning system according to claim 7, characterized in that, include: The fault prediction module is also used to process the power curve data of the degradation state using the sliding window method, and divide it into several data sequences, the data sequences including a training set; The fault prediction module is also used to explore the potential relationships between the data in the training set using a neural network model, and to predict the power curve data of the degradation state in the next working cycle.

11. The early warning system according to claim 7, characterized in that, include: The fault prediction module is also used to divide the data sequence into sample sets, including a validation set and a test set; The fault prediction module is also used to adjust and determine the optimal hyperparameters of the neural network model using the validation set; The fault prediction module is also used to evaluate the performance of the neural network model after the hyperparameters have been determined using the test set.

12. The early warning system according to claim 8 or 9, characterized in that, include: The fault warning module is also used to obtain the degradation state classification with the highest similarity to the predicted degradation state classification, and the corresponding working state type is used as the predicted turnout working state. The fault warning module is also used to determine the fault type with the highest similarity to the predicted degradation state classification, and to issue a fault warning.

13. An electronic device, characterized in that, include: processor; And a memory storing computer-executable instructions, which, when executed, cause the processor to perform the method according to any one of claims 1-6.

14. A computer storage medium, characterized in that, in, The computer storage medium stores one or more programs that, when executed by a processor, implement the method of any one of claims 1-6.

Images