A Smart Fault Diagnosis Method for Railway Centralized Dispatch Systems

Abstract

An intelligent fault diagnosis method for centralized railway dispatching systems includes: collecting information from three dimensions—single device nodes, network communication traffic, and system functions—and designing fault detection methods for each dimension to extract system fault characteristics; performing root cause analysis of system faults; and forming a historical decision knowledge base based on historical knowledge and mined association rules to automatically match handling suggestions. This invention improves the efficiency of maintenance personnel manually analyzing system data; it analyzes multi-dimensional, heterogeneous, and massive data using data mining methods to automatically discover potential and unknown correlations between system characteristics and fault root causes, providing strong support for CTC system maintenance and fault recovery; it realizes the transformation from "fault repair" to "preventive repair"; and it combines fault rules obtained from data mining with historical experience to form a fault diagnosis rule base, assisting maintenance personnel in accurately locating problem nodes, quickly handling faults, and improving maintenance efficiency.

Description

Technical Field

[0001] This invention relates to a fault diagnosis method, and more particularly to an intelligent fault diagnosis method for a centralized railway dispatching system. Background Technology

[0002] As the railway network continues to expand and its structure becomes increasingly complex, the requirements for the safe and stable operation of train dispatching and command systems are also rising. The Centralized Traffic Control (CTC) system, as the command center of high-speed railway transportation, integrates technologies from multiple disciplines, including communication, signaling, transportation organization, and networking. While its complex system structure ensures the realization of diverse functions, the possibility of failure is also constantly increasing. In addition to CTC system failures caused by hardware malfunctions, complex problems caused by network communication delays, software logic defects, and system data errors are becoming increasingly prominent. Timely and accurate fault detection and root cause location are fundamental and crucial for reducing the impact of system failures and ensuring efficient train operation. Traditional fault diagnosis techniques primarily target hardware equipment failures, and system maintenance relies on manual methods such as equipment inspection, log analysis, and network testing, making it difficult to achieve intelligent identification and rapid recovery of system anomalies. How to automatically detect system faults and accurately analyze their causes through multi-dimensional data correlation analysis, thereby providing auxiliary decision-making and improving maintenance efficiency, is a pressing issue that needs to be addressed.

[0003] With the development of emerging technologies such as big data, artificial intelligence, and machine learning, correlation analysis using multi-source and multi-dimensional data has become a research hotspot. Applying technologies such as data fusion, deep learning, and text mining to railway system operation and maintenance is conducive to forming a new intelligent operation and maintenance model for high-speed railway CTC systems that is highly reliable, efficient, and low-cost. Therefore, the intelligent fault diagnosis solution is based on multi-source heterogeneous data and designs detection methods around three levels: equipment nodes, network communication, and system functions. By integrating multi-dimensional data such as equipment parameters, communication status, and fault text, it deeply mines the correlation rules and hidden patterns between system characteristics and fault root causes, promptly detects system faults, automatically analyzes the causes of faults, and provides fault recovery suggestions, thereby improving the intelligence level of CTC system operation and maintenance and reducing operation and maintenance pressure.

[0004] Existing fault detection methods for railway signaling systems are mostly single-dimensional, falling into two main categories: fault diagnosis based on equipment electrical characteristics and anomaly analysis based on the overall system structure. The former primarily focuses on track circuits and turnout switching machines, employing methods such as circuit simulation and knowledge graphs for fault diagnosis, resulting in relatively singular detection targets and data sources. The latter, on the other hand, focuses on the railway signaling system itself, using methods such as decision trees and graph theory reasoning to perform retrospective analysis after a system fault occurs, lacking state assessment and anomaly early warning. Summary of the Invention

[0005] Current fault detection and operation maintenance of CTC systems heavily rely on expert experience, operational guidelines, and fragmented fault analysis records, making accurate fault detection and rapid root cause analysis difficult. To improve maintenance efficiency and reduce the impact of faults, this invention designs an intelligent fault diagnosis method for railway centralized dispatching systems, with the following steps:

[0006] Step 1: Collect information from three dimensions: single device node, network communication traffic, and system function, and design fault detection methods for each dimension to extract system fault characteristics;

[0007] Step 2: Use association rule algorithms to fuse multi-dimensional features to achieve root cause analysis of system failures;

[0008] Step 3: Based on historical knowledge and the association rules obtained through mining, a historical decision-making knowledge base is formed, and disposal suggestions are automatically matched.

[0009] Beneficial effects

[0010] 1. By extracting fault characteristic data of CTC system from three dimensions—single device node, network communication traffic, and system function—the diagnostic results are more accurate compared to single-dimensional fault diagnosis methods, providing strong support for CTC system maintenance and fault recovery.

[0011] 2. Given the large volume of CTC system log data, the FP-growth algorithm is selected when using multi-dimensional fault feature data for correlation analysis. This approach can achieve the goal of generating system fault correlation rules while reducing computational overhead and improving fault analysis efficiency.

[0012] 3. Combine fault rules with historical experience to form a historical fault diagnosis rule library, which helps maintenance personnel accurately locate problem nodes, quickly determine diagnostic conclusions, and guide on-site troubleshooting. Attached Figure Description

[0013] Figure 1 This is a diagram of the CTC system fault detection architecture.

[0014] Figure 2 Flowchart for implementing the system fault diagnosis rule base. Detailed Implementation

[0015] Currently, CTC systems generally adopt the principle of intelligent decentralized autonomous design, applying distributed computing and control technologies to achieve unified command of train and shunting operations. The overall network structure of the CTC system is divided into three layers: the China Railway Corporation dispatching system, the railway bureau central subsystem, and the station subsystem. Among them, the central subsystem is generally located in the railway bureau dispatching center and is responsible for controlling the operation of trains in the entire dispatching section; the station subsystem is used to receive train dispatching instructions, issue them to the interlocking system in a timely manner, and convert the dispatching instructions into equipment execution commands. The main hardware equipment, network transmission, and software applications of the two are shown in Table 1.

[0016] Table 1. Main Components of CTC Center and Station Subsystems

[0017]

[0018]

[0019] Data statistics show that station subsystem failures account for over 95% of all failures. Therefore, taking the station's CTC system as an example, the fault detection system architecture is designed as follows: Figure 1 The system collects operational status and related parameters from three dimensions: "device nodes, communication transmission, and application functions." Fault detection algorithms are designed based on the system's operating mechanism to extract abnormal system features. CTC system fault state simulations are performed, and system logs are used to reconstruct the data characteristics of each dimension under fault conditions and their corresponding root causes, forming a training dataset. Hidden relationships between fault features and root causes are mined using association rule algorithms, and the accuracy of fault detection and root cause analysis is analyzed through simulation experiments.

[0020] This invention discloses an intelligent fault diagnosis method for a centralized railway dispatching system, the steps of which are as follows:

[0021] Step 1: Collect information from three dimensions: single device node, network communication traffic, and system function, and design fault detection methods for each dimension to extract system fault characteristics;

[0022] Fault Feature Extraction

[0023] As the informatization level of the CTC system continues to improve, the data it accumulates has also grown explosively, forming a massive dataset containing a large amount of multi-source heterogeneous data, including ontological data, monitoring data, patrol data, and external data, providing a foundation for the application of technologies such as data mining and artificial intelligence. However, the current CTC system data still has significant problems such as non-standard data formats, unbalanced sample data, coarse classification rules, and difficulties in data fusion. To improve data quality, optimize data structure, extract key information, and design corresponding fault feature extraction methods for data of different dimensions, this paper proposes to promptly detect system anomalies and provide a data foundation for subsequent root cause analysis.

[0024] (1) Device node feature extraction

[0025] Extract the characteristics of the system hardware devices, and output the faulty device's number, name, and abnormal data category as fault characteristics.

[0026] First, it detects the status of equipment, automatically collects and dynamically updates the configuration information of the main equipment in the CTC system, such as equipment model, manufacturer, online time, installation cabinet location, power network wiring terminals, etc. If the configuration information cannot be obtained or is inconsistent with the information in the list, it is determined to be a component failure.

[0027] Second, load parameters are monitored. Normal thresholds for load parameters of the monitored objects are set according to equipment type and historical operating data, including equipment system time deviation, number of equipment processes, and equipment resource utilization. When the monitored parameters exceed the threshold range, it is identified as an equipment fault.

[0028] (2) Network communication feature extraction

[0029] Test network connectivity and count network traffic. Identify network nodes that are disconnected or have abnormal traffic trends as fault nodes and output their node names and anomaly types as fault characteristics.

[0030] First, by testing the connectivity of each network device, the network topology is discovered. The online status and port connection status of devices are detected through modes such as SNMP, ICMP, and ARP. A network topology map is then drawn, and the status of all network devices and abnormal nodes are located through the topology map.

[0031] Secondly, by monitoring network traffic in real time, traffic monitoring and trend analysis are achieved to quickly identify and extract abnormal network communication characteristics. The internal communication traffic and external interface traffic of the CTC system are time series that fluctuate to a certain extent depending on the device nodes, service requirements, and actual line operation. If a fixed threshold is set directly based on historical experience, an excessively large threshold range will fail to detect traffic anomalies in a timely manner; an excessively small threshold range will easily generate false alarms and reduce detection accuracy. This paper uses statistical process control charts to monitor the traffic time series and automatically adjusts the detection threshold according to the actual business situation of the system, thereby accurately distinguishing between normal traffic fluctuations and abnormal changes.

[0032] Statistical process control charts (SPCs) can design control limits based on the statistical characteristics of observed targets to determine whether anomalies have occurred in the production process. The exponentially weighted moving average (EWMA) control chart algorithm, which uses an exponentially decreasing weighted moving average, is one of the representative algorithms of SPCs. EWMA can detect changes in variance and is sensitive to mean drift. It can combine historical statistical knowledge to perform anomaly analysis on current data, effectively detecting subtle changes in the data.

[0033] Taking the communication traffic between the station's automated control unit and the station's exchange as an example, the observation x(i) represents the traffic volume at different times i. For a set of observations x(i), the geometric moving average z(i) is used to plot the EWMA control chart.

[0034] z(i)=λ·x(i)+(1-λ)·z(i-1),0<λ≤1

[0035] Where λ is the smoothing coefficient, used to adjust the sensitivity of z(i) to x(i).

[0036] The threshold range for anomaly detection can be calculated using the mean and variance of the predicted values ​​z(i). The mean μ z and variance σ z satisfy

[0037] μ z =μ x

[0038]

[0039] Where, μ x Let σ be the mean of the historical observation sequence {x(1),x(2),...,x(i-1)}. x Let U be the variance of the observed sequence. Then the upper boundary U of the EWMA control chart... z and lower boundary D z Together, they form the threshold range for traffic anomaly detection, and their calculation formulas are as follows:

[0040] U z =μ z +L·σ z

[0041] D z =μ z -L·σ z

[0042] Where L is the boundary coefficient.

[0043] After calculating the upper and lower boundaries of the EWMA control chart from the historical flow observation sequence {x(1),x(2),...,x(i-1)}, it is determined whether the current flow prediction value z(i) exceeds the normal flow threshold range [U z D z This allows us to determine if there are any abnormal traffic patterns.

[0044] (3) System functional fault extraction

[0045] During the operation of the CTC system, a large number of system logs described in text form are retained, recording a wealth of real and valuable information such as system status, user operations, interface information, and error reports. Effective text mining technology can automatically extract key information from the logs, quickly identify system faults, and provide reliable training data for root cause analysis.

[0046] System logs record system operation information in real time and comprehensively. They are massive in volume, highly repetitive, and exhibit a significant imbalance between normal and fault data. Directly analyzing all system logs not only consumes substantial computational resources but also affects the accuracy of the data mining results. To achieve efficient analysis and accurate identification of system anomalies, this paper uses CTC system logs under fault conditions as the data source for text mining and feature extraction. The data characteristics and required preprocessing operations are as follows:

[0047] ① Data Balance. Because the frequency, record length, symptom description, and solutions for different faults vary within the system—for example, software configuration errors and dispatcher mistakes are relatively frequent, while channel interference and acquisition board malfunctions occur less often—directly using raw data can easily lead to inaccurate classification. Therefore, the collected system logs need to be manually filtered and supplemented to ensure a relatively balanced distribution of different data types.

[0048] ② Text Cleaning. The CTC system logs contain a large amount of noisy data such as time, train number, and section name, which is detrimental to the extraction of effective features. Daily master data needs to be cleaned; this paper achieves this by writing regular expressions. For example, the regular expression for extracting train numbers is as follows:

[0049] [KZTXGDC]\d{1,5}[times]||\d{1,5}[times]

[0050] This expression means: match a string that begins with "K" or "Z" or "T" or "X" or "G" or "D" or "C", followed by one to five digits, and then "times" (e.g., "K12345 times" or "Z23 times"); or match a string that begins with one to five digits, followed by "times" (e.g., "12345 times" or "3 times").

[0051] ③ Text segmentation. The CTC system log contains a large number of railway-related professional terms. To achieve accurate segmentation, this paper uses the Jieba segmentation tool. In addition to using a general dictionary, a custom railway signaling professional dictionary is added, including line and station names, train operation and dispatching professional terms, and names of various equipment.

[0052] After data preprocessing, fault text features are extracted and vectorized based on TF-IDF (Term Frequency-Inverse Document Frequency). TF-IDF is a weighted term frequency statistical method used to characterize the importance of words in text. Its principle is: when a word appears frequently in one text but less frequently in other texts, it is considered a keyword in that text. TF-IDF consists of two parts: term frequency and inverse document frequency. i,j for

[0053]

[0054] In the formula, n i,j For the word w i In text d j The number of times it appears in; For text d j The sum of the occurrences of all words in the fraction; add 1 to the denominator to avoid the denominator being 0.

[0055] Reverse Document Frequency (IDF) i for

[0056]

[0057] In the formula, k i For the word w i The number of related texts in the text set D; N is the size of D; add 1 to the denominator to avoid the case where the denominator is 0.

[0058] Combining the two, we get text d. j Chinese word w i weight W i,j for

[0059] W i,j =TF i ×IDF i

[0060] Then text d j The vector representation can be constructed using word weights.

[0061] d j =[W 1,j W 2,j ,…,W n,j ]

[0062] Text data vectorized by TF-IDF is characterized by sparsity and high dimensionality, and the resulting word weights are continuous values. Directly using these values ​​for subsequent association rule mining is inefficient. Therefore, the weight values ​​need to be discretized, and important feature attributes need to be extracted. Based on field experience and expert advice, a threshold is set for the feature word weights to classify their importance into three categories: low (0), moderate (1), and high (2). When W... i,j When W < 0.01, the importance is low; when 0.01 ≤ W i,j When W is ≤0.03, the importance is average; when W i,j >0.03 indicates high importance. Feature words with moderate and high importance are retained as text keywords for subsequent association rule mining, such as acquisition board malfunction, train schedule window delay, communication interruption, and control command issuance failure.

[0063] Step 2: Use association rule algorithms to fuse multi-dimensional features to achieve root cause analysis and mining of system faults;

[0064] Root cause analysis

[0065] Due to the diversity and complexity of CTC system functions, anomalies in one or more factors can affect the state of related dimensions, gradually spreading and leading to the accumulation and propagation of risks, ultimately resulting in system failure. Therefore, relying on anomaly detection in a single dimension is insufficient to determine the root cause of a failure. It is necessary to explore the correlation between multi-dimensional anomaly characteristics and the true cause of the failure, thereby achieving fault location, root cause analysis, and decision support.

[0066] Association rule mining can extract correlations between different itemsets by mining all frequent itemsets in a database whose support and confidence exceed a specified threshold. It has been widely applied in mobile communications, network security, and fault diagnosis. The Apriori algorithm is the most commonly used method for mining frequent itemsets, offering advantages such as simplicity, ease of understanding, and low data requirements. However, due to the large data scale of the CTC system, applying the Apriori algorithm requires continuous scanning of the dataset, leading to excessively long computation times. To improve computational efficiency, this paper adopts the FP-Growth algorithm for frequent itemset mining. This algorithm uses a frequent pattern tree (FP-tree) to compress frequent itemsets, dividing the compressed database into a set of conditional datasets, and then mining association rules from each conditional dataset. The FP-Growth algorithm does not generate candidate itemsets but directly traverses all frequent itemsets on the FP-tree, thus greatly improving algorithm performance.

[0067] Taking the common train occupancy loss in the CTC system as an example, we will conduct root cause analysis. Let I = {i1, i2, ..., i mGiven an itemset containing m items, extract the system logs and fault reports related to train occupancy loss faults at a certain station over the past three years to form a transactional database D = {r1, r2, ..., r...}. m}, each transaction r i Each (i = 1, 2, ..., n) corresponds to a subset of I, satisfying ... I is the set of all data items in D.

[0068] Let X and Y be itemsets, and The association rule is then represented as

[0069]

[0070] In the formula, X represents the prerequisite, and Y represents the association result. When itemset X occurs, itemset Y will also occur with a certain probability. For example, if the primary and backup machine information synchronization fails, the database server may be malfunctioning.

[0071] The parameters used to evaluate association rules are support and confidence. Support is the primary condition for discovering association rules; it refers to the probability that a transaction {X, Y} exists in the database D, i.e., the probability that the entire itemset contains both X and Y.

[0072]

[0073] Support filters prevalent association rules from the "quantity" dimension. Based on the database structure characteristics and anomaly detection needs, a minimum support s is set, and rules with low frequency of occurrence are eliminated. The remaining non-empty subset is called the frequent itemset.

[0074] Confidence score is a secondary condition for discovering association rules; it refers to the probability that Y will occur given that the prerequisite X has occurred.

[0075]

[0076] The confidence level filters the required association rules from the "quality" dimension, reflecting the reliability of the association rules. A minimum confidence level *c* is set to filter out strongly associated rules.

[0077] The association rule algorithm of this invention adopts the FP-growth algorithm, and its operation steps are as follows:

[0078] Step 1: Calculate the number of repetitions for each item and sort them sequentially. Build the FP-tree for this transaction based on the frequent itemset in descending order.

[0079] Step 2: Calculate the minimum support, delete items with less than the minimum support, and create a header table of remaining items;

[0080] Step 3: Access the FP-tree from bottom to top according to the item header table to generate the conditional schema base;

[0081] Step 4: Based on the conditional pattern base, re-determine whether the conditional FP-tree is empty or contains only a single path. If it is, proceed to Step 5; otherwise, go to Step 3.

[0082] Step 5 combines the nodes in the path and generates the final frequent pattern with the frequent suffix items.

[0083] In train occupancy loss faults, the possible fault characteristics and root cause types are first identified and numbered. The fault characteristics are divided into three dimensions: equipment nodes, communication transmission, and application functions. The characteristics of the first two dimensions are directly extracted from the system operation data, while the characteristics of the application function dimension are extracted from the system logs.

[0084] The equipment nodes include switch A, switch B, autonomous machine A, autonomous machine B, vehicle service terminal A, vehicle service terminal B, integrated maintenance terminal, and electrical maintenance terminal. Regarding equipment status, D1, D2, D3, D4, D5, D6, D7, and D8 represent the status of the corresponding equipment nodes, with values ​​ranging from 0 (normal) to 1 (abnormal). Regarding load parameters, D9, D10, D11, D12, D13, D14, D15, and D16 represent the compliance parameters of the corresponding equipment nodes, with values ​​ranging from 0 (normal), 1 (below threshold), and 2 (above threshold).

[0085] Communication transmission includes automatic control machine A, automatic control machine B, train operation terminal A, train operation terminal B, integrated maintenance terminal, and electrical maintenance terminal connected to the station's Ethernet network, as well as serial communication between the station's automatic control machines and the interlocking system, and Ethernet communication between the station's automatic control machines and the train control system. Regarding connectivity status, T1, T2, T3, T4, T5, T6, T7, and T8 represent the connectivity status of the corresponding communication transmission, with values ​​ranging from 0 (connected) to 1 (disconnected). Regarding communication traffic, T9, T10, T11, T12, T13, T14, T15, and T16 represent the data traffic of the corresponding communication transmission, with values ​​ranging from 0 (normal), 1 (below threshold), and 2 (above threshold).

[0086] The system's functional implementation status cannot be directly extracted from the system; text mining of the system logs is required to extract the following fault feature table:

[0087] Table 2. Fault Characteristics Related to Train Occupancy Loss in Application Function Dimensions

[0088]

[0089]

[0090] In addition to fault characteristics across various dimensions, to achieve association rule mining, it is also necessary to clarify the causes of system faults, including equipment failures, communication failures, data errors, operational mistakes, etc. Based on expert experience and fault log extraction results, the fault cause table is as follows:

[0091] Table 3. Root causes of train occupancy loss.

[0092]

[0093] 10,000 data points were randomly selected from the database for the experiment. The data were encoded and assigned values ​​according to the above rules, with 75% used as the training set and 25% as the test set. Rule mining was performed according to the FP-growth algorithm. With a minimum support of 0.02, 2276 association rules were obtained; with a minimum confidence of 0.6, 1568 strong association rules were obtained. Some of the association rules are shown in Table 4.

[0094] Table 4. Train Occupancy Loss Fault Association Rules (Partial)

[0095]

[0096] Step 3: Based on historical knowledge and the association rules obtained through mining, a historical decision-making knowledge base is formed, and disposal suggestions are automatically matched.

[0097] Decision support generation

[0098] The detection features extracted from three dimensions—single device node, network communication traffic, and system functional failure—are used to generate diagnostic rules using the FP-growth algorithm. The generated diagnostic rules are then checked to see if they already exist in the historical fault diagnosis rule base. If the rule exists, the diagnostic conclusion is determined directly based on the fault cause associated with the rule and its corresponding handling suggestions, guiding on-site troubleshooting. If the rule does not exist, it is added to the fault diagnosis rule base, and the fault cause and handling measures are recorded in the historical decision knowledge base.

[0099] This invention collects multi-dimensional data from the CTC system and designs corresponding methods to extract system fault features based on the characteristics of different data dimensions. Using fault causes as the mining target, it uncovers potential and unknown correlations between fault features and root causes, thereby enabling timely alarms and automatic fault cause analysis when system anomalies are detected. The mined stable fault cause rules are incorporated into fault diagnosis rules, providing strong support for maintenance personnel in system fault analysis, cause determination, and troubleshooting.

[0100] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. An intelligent fault diagnosis method for railway centralized dispatching systems, characterized by: Step 1: Collect information from three dimensions: single device node, network communication traffic, and system function; design fault detection methods for each dimension; and extract system fault characteristics. Network communication traffic extraction: By testing the connectivity of each network device, the network topology is discovered. The online status of devices and port connection status are detected through SNMP, ICMP, and ARP modes. A network topology map is drawn, and the status of all network devices and abnormal nodes are queried and located through the topology map. By detecting network traffic in real time, traffic monitoring and trend analysis are achieved, and abnormal network communication characteristics are quickly identified and extracted. Step 2: Use association rule algorithms to fuse multi-dimensional features to achieve root cause analysis of system failures; Taking train occupancy loss in the CTC system as an example for fault root cause analysis: Let I = {i1, i2, ..., i m Given an itemset containing m items, extract the system logs and fault reports related to train occupancy loss faults at a certain station over the past three years to form a transactional database D = {r1, r2, ..., r...}. m }, each transaction r i Each (i = 1, 2, ..., n) corresponds to a subset of I, satisfying ... I is the set of all data items in D; Let X and Y be itemsets, and , , The association rule is represented as: X is the prerequisite, and Y is the associated result; when itemset X occurs, itemset Y occurs with a certain probability. The parameters for evaluating association rules are support and confidence; support refers to the probability that a transaction {X,Y} will occur in database D, that is, the probability that all itemsets contain both X and Y. ； Support filters prevalent association rules from a quantitative perspective. A minimum support 's' is set based on database structure characteristics and anomaly detection needs. Rules with low frequency are removed, and the remaining non-empty subset is called the frequent itemset. Confidence refers to the probability that Y will occur given the precondition X. ； The confidence level is used to filter the required association rules from the quality dimension, reflecting the reliability of the association rules. A minimum confidence level c is set to filter out strong association rules. Step 3: Based on historical knowledge and the association rules obtained through mining, a historical decision-making knowledge base is formed to automatically match disposal suggestions; the association rule algorithm adopts the FP-growth algorithm; the feature extraction of a single device node includes the following: the FP-growth algorithm, whose operation steps are as follows: Step 1: Calculate the number of repetitions for each item and sort them sequentially. Build the FP-tree for this transaction based on the frequent itemset in descending order. Step 2: Calculate the minimum support, delete items with less than the minimum support, and create a header table of remaining items; Step 3: Access the FP-tree from bottom to top according to the item header table to generate the conditional schema base; Step 4: Based on the conditional pattern base, re-determine whether the conditional FP-tree is empty or contains only a single path. If it is, proceed to Step 5; otherwise, go to Step 3. Step 5 combines the nodes in the path and generates the final frequent pattern with the frequent suffix items; The extraction of system faults includes the following: text mining and feature extraction are performed using CTC system logs under fault conditions as the data source. The data characteristics and required preprocessing operations are as follows: ① Data Balance: Since the frequency of occurrence, record length, phenomenon description and solution of different faults in the system are different, it is necessary to manually screen and supplement the collected system logs to make the different types of data relatively balanced; ② Text cleaning: The CTC system log contains a large amount of noisy data such as time, train number, and section name, which is not conducive to the extraction of effective features. The log data needs to be cleaned, which can be achieved by writing regular expressions. ③ Text segmentation: The CTC system log contains a large number of railway-related professional terms. To achieve accurate segmentation, the Jieba segmentation tool is used. Based on the general dictionary, additional terms are added, including the names of railway stations, train dispatching and control professional terms, and custom railway signals for various equipment. After data preprocessing, fault text features are extracted and vectorized based on TF-IDF. The text data vectorized by TF-IDF is characterized by sparsity and high dimensionality, and the obtained word weights are continuous values. Directly using them for subsequent association rule mining is inefficient. It is necessary to discretize the weight values ​​and extract important feature attributes. Based on on-site experience and expert advice, the importance of feature words was divided into three categories: low (0), moderate (1), and high (2) by setting thresholds for their weights. Feature words with moderate and high importance were retained as text keywords for subsequent association rule mining. Step 3 includes the following: Detection features extracted from three dimensions—single device node, network communication traffic, and system functional faults—are used to generate diagnostic rules using the FP-growth algorithm. The generated diagnostic rules are then checked to see if they already exist in the historical fault diagnosis rule base. If the rule exists, the diagnostic conclusion is determined directly based on the fault cause associated with the rule and its corresponding handling suggestions, guiding on-site troubleshooting. If the rule does not exist, it is added to the fault diagnosis rule base, and the fault cause and handling measures are recorded in the historical decision knowledge base.

2. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute the method of claim 1.

3. An electronic device, characterized in that, It includes a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions, when executed, perform the method of claim 1.

Images