Navigation equipment fault anomaly detection method based on multi-layer dynamic causal relationship network

Abstract

This invention relates to a method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network. It involves collecting multi-source data from the general aviation system and constructing a multi-layer dynamic causal relationship network encompassing both a system macro-level and an equipment level. A message queue is used to process the real-time data stream of the general aviation system, extracting feature data for real-time fault detection. The multi-layer dynamic causal relationship network is then used to analyze the feature data to determine if any equipment anomalies exist. If so, the multi-layer dynamic causal relationship network is further used to locate the cause of the equipment anomaly. If an equipment anomaly is found, a fault alarm is issued. This invention achieves real-time automatic detection and location of equipment faults, facilitating engineers in handling equipment faults, improving fault handling efficiency, and effectively ensuring the safety and stability of general aviation.

Description

Technical Field

[0001] This invention belongs to the field of data processing and fault detection technology, specifically relating to a method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network. Background Technology

[0002] Currently, the demand for shipping across the country is constantly growing, and the number of dams and locks with navigation functions is increasing daily. As an important navigation dam in the Yangtze River basin, the Three Gorges Dam handles approximately 100 vessels daily. This navigation scenario involves locks, control centers, hydrological monitoring equipment, and communication equipment. The complex and massive navigation system requires a high degree of interdependence and coordination among various devices to complete its navigation tasks, thus inevitably leading to malfunctions. Rapid fault detection and localization are urgently needed solutions.

[0003] Traditional fault detection methods rely primarily on manual or partial manual labor, but these are insufficient to meet the demands of today's diverse equipment and intensive shipping industry. A single malfunction in navigation equipment can lead to system shutdowns or severe economic losses; for example, a lock malfunction can cause vessel delays, disrupt shipping plans, and affect the operation of locks at different levels. In recent years, the Three Gorges Navigation Administration has established CCTV monitoring systems, network monitoring systems, and dispatching systems to achieve digital navigation management and facilitate efficient and safe navigation operations. These systems are used for real-time monitoring and management of navigation equipment. While these systems can capture a large amount of operational data from navigation equipment in real time, they lack effective analysis methods, making it difficult to extract key information and perform other tasks. Therefore, it is imperative to provide a new intelligent fault detection method that fully utilizes the aforementioned multi-source data and combines it with dynamic causal network analysis.

[0004] Existing fault detection technologies primarily rely on pre-defined rules and simple data analysis methods. In the Three Gorges navigation scenario, many fault detection methods still depend on fault alarm rules provided by equipment manufacturers, such as those based on operating thresholds for equipment current, water level, and container pressure. An alarm is triggered only when these thresholds are exceeded. However, this threshold-based approach is overly reliant on manual settings and cannot handle the complex behavior of equipment under different operating conditions, easily leading to false alarms and missed alarms. Furthermore, certain monitoring data cannot accurately pinpoint the faulty component. In addition, fault detection methods based on statistical analysis, such as simple statistical indicators like averages and standard deviations, while capable of detecting abnormal changes in some equipment, struggle to capture complex dynamic changes in multi-source data. Especially when dealing with the diverse data types involved in navigation equipment, these methods demonstrate insufficient comprehensiveness in their detection effectiveness.

[0005] Existing fault location methods typically rely on alarm signals provided by the aforementioned fault detection systems. They then infer the location of the fault by analyzing the operational logs of the general aviation equipment servers, monitoring data, and the experience of technical personnel. Fault location techniques generally require experienced technicians to investigate step-by-step. For example, when an abnormal alarm appears in the general aviation declaration database, technicians first check the alarm content and then examine the status of each server in the general aviation declaration system. When high CPU usage is detected on the general aviation management server, relevant CPU metrics are checked one by one to further pinpoint the problem. Through these checks, the conclusion is ultimately reached that the issue was caused by a resource-intensive query requested by the general aviation control center. In addition, technicians need to record the problem details and generate reports. This fault location technique requires significant human resources and time. In complex general aviation systems, the required human resources may be insufficient to support fault monitoring and location in the event of a large-scale equipment failure, leading to downtime or severe economic losses.

[0006] In recent years, machine learning algorithms have been increasingly applied to equipment fault detection. These algorithms are trained on equipment operating data using supervised or unsupervised learning methods to learn the data characteristics of different fault types, and are then used to detect faults once they mature. However, the application of these methods faces challenges such as difficulties in data annotation, long model training times, and the difficulty in interpreting complex faults. Due to the dynamic nature of the operating environment of general aviation equipment, there are potential dynamic causal relationships between multi-source data of the equipment, and existing algorithms have difficulty handling dynamic causal relationships. Therefore, the accuracy and real-time performance of fault detection still need to be further improved. Summary of the Invention

[0007] The purpose of this invention is to address the aforementioned problems by providing a method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network. This method constructs a multi-layer dynamic causal relationship network, including a system macro-level and an equipment level, using the NOTEARS algorithm. The dynamic causal network at the system macro-level identifies dependencies between various general aviation subsystems, while the dynamic causal network at the equipment level identifies potential fault chains between devices. By combining distributed stream processing technology and a real-time message transmission mechanism using message queues, the method performs real-time analysis of multi-source data from the general aviation system, enabling real-time detection and location of equipment faults. This facilitates equipment fault handling by engineers, improves fault handling efficiency, and ensures the safety and stability of general aviation.

[0008] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0009] A method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network includes the following steps:

[0010] Step S1: Collect multi-source data from the navigation system. The multi-source data includes navigation equipment data, navigation equipment server logs, navigation monitoring data, navigation execution work orders, and navigation vessel information data.

[0011] Step S2: Preprocess the multi-source data, including data standardization, conversion into time series, and data timestamp synchronization;

[0012] Step S3: Construct a multi-layer dynamic causal relationship network, including constructing a dynamic causal network at the system macro level and a dynamic causal network at the equipment level; the dynamic causal network at the system macro level is used to identify the dependencies between various general aviation subsystems, and the dynamic causal network at the equipment level is used to handle the nonlinear dynamic causal relationships of general aviation equipment and identify potential fault chains between equipment.

[0013] Step S4: Use a message queue to process the real-time data stream of the general aviation system and extract the feature data for real-time fault detection.

[0014] Step S5: Use the dynamic causal relationship network obtained in step S3 to analyze the feature data obtained in step S4 to determine whether there is a device malfunction. If the determination result is yes, then further use the dynamic causal relationship network to locate the cause of the device malfunction.

[0015] Step S6: If there is a device malfunction based on the judgment result of step S5, a fault alarm will be issued.

[0016] Furthermore, step S2 specifically includes the following sub-steps:

[0017] S201: Use the z-score method to standardize the indicator data of different devices;

[0018] S202: Use the three-standard-deviation method to remove outliers and ensure data accuracy;

[0019] S203: Use the drain algorithm to extract the log data of the general aviation equipment server into events and convert it into a time series; use the TraceExtract algorithm to extract the critical path of the call chain data and convert it into a time series.

[0020] S204: Timestamp synchronization of data for each category enables multivariate time series analysis of general aviation equipment indicators, work orders, server logs, general aviation monitoring and asset data.

[0021] Preferably, step S3 uses an improved NOTEARS algorithm to construct a multi-layer dynamic causal network. The improvement of the existing NOTEARS algorithm by the improved NOTEARS algorithm includes the following steps:

[0022] 1) Time series data modeling: The core of dynamic causal relationship modeling lies in processing data with changing time dimensions and learning the causal structure that changes over time on the time axis; dividing time series data into time slices, with each time slice corresponding to observations over a certain period of time; introducing past observations as lagged variables and establishing the causal relationship between lagged variables and current variables;

[0023] 2) Extend the graph structure of the NOTEARS algorithm to a temporal structure. In the original optimization problem of the NOTEARS algorithm, the dynamic causal relationship is a static graph. Change the edges in the static graph to time-dependent edges to obtain a dynamic graph. Accordingly, add time dimension constraints to ensure the acyclicity of the dynamic causal graph. Add cross-time acyclicity constraints to the original NOTEARS algorithm constraints.

[0024] The Notears algorithm is used to learn the dynamic causal relationships within each time slice, and then a dynamic Bayesian network is used to add directed edges between time slices. These directed edges are used to represent the dependencies between lagged variables and current variables.

[0025] 3) Dynamic optimization: For each time slice, the original optimization strategy of Notears is still used, namely minimizing the squared error or other loss functions.

[0026] Preferably, step 3 specifically includes the following sub-steps:

[0027] Step S301: Feature selection and data transformation;

[0028] At the macro level of the general aviation system, based on business knowledge and data exploration results, time slices are divided according to the time dimension, and key features at the level of general aviation subsystems are selected within each time slice; the feature data of each general aviation subsystem is represented as a feature matrix X. (t) _subsystems;

[0029] At the equipment level within the general aviation subsystem, for each general aviation subsystem, time slices are divided according to the time dimension, and relevant data of its internal equipment are selected within each time slice; a feature matrix X at the equipment level is constructed for each general aviation subsystem. (t) _devices;

[0030] Step S302: Initialize the weighted adjacency matrix and Lagrange multipliers of the dynamic causal network, and construct the smoothing function;

[0031] At the macro level of the general aviation system, an initial weighted adjacency matrix W is selected for the optimization problem of the improved NOTEARS algorithm. (t) 0_subsystems and Lagrange multiplier α0; constructing the system-level intra-time slice smoothing function h(W(t) a) Enables it to encode acyclic constraints; constructs acyclic constraint function ht(W) between system-level time slices. (t) a) is used to ensure the acyclicity constraint of dynamic causal graphs from the time dimension;

[0032] At the device level within the general aviation subsystem, initialize the device-level weighted adjacency matrix W. (t) 0_devices and Lagrange multiplier β0; construct the device-level time-slice smoothing function h(W (t) b) Enables it to encode acyclic constraints; constructs acyclic constraint function ht(W) between device-level time slices. (t) b) is used to ensure the acyclicity constraint of the dynamic causal graph from the time dimension;

[0033] Step S303: Optimize the weighted adjacency matrix to obtain a dynamic directed acyclic graph, i.e., a dynamic causal network;

[0034] Numerical optimization methods are used to minimize the NOTEARS objective function at both the macroscopic level and the equipment level within subsystems of the general aviation system, satisfying h(W (t) a)=0、h(W (t) b)=0 intra-time slice acyclic constraint and time-dimension acyclic constraint ht(W) (t) a)=0、ht(W (t) b)=0; simultaneously iteratively update the system-level weighted adjacency matrix W. (t) _devices and the device-level weighted adjacency matrix W (t) _subsystems, will optimize the matrix W (t) _devices and W (t) After thresholding, the subsystems are transformed into a dynamic directed acyclic graph, which serves as a dynamic causal network.

[0035] Step S304: Verify and correct the dynamic causal network;

[0036] At the macro level of the general aviation system, test data from the general aviation subsystems are used to verify the learned dynamic causal network between the general aviation subsystems. If an incorrect causal chain is found, manual intervention and correction are carried out on the incorrect causal chain to finally determine the system-level dynamic causal network.

[0037] At the equipment level within the general aviation subsystem, equipment-level test data is used to verify the dynamic causal network between devices within each general aviation subsystem. Through manual intervention and correction of unreasonable dynamic causal relationship chains, the dynamic causal network structure at the equipment level is finally determined.

[0038] Preferably, in step S4, the real-time data stream of the aviation system is transmitted using the message queue Kafka, the real-time data stream is processed using the stream processing engine Flink, and the feature data in the real-time data stream is analyzed and extracted using a sliding window.

[0039] Furthermore, in step S5, based on the multi-layer dynamic causal relationship network obtained in step S3, the feature data of the real-time data stream provided by the Flink streaming engine is analyzed, the health score of the device is calculated according to the feature data, and compared with a predetermined threshold to determine whether the device status is abnormal. The causal chain of the device abnormality is traced through the directed acyclic graph of the multi-layer dynamic causal network to find the root cause of the device abnormality.

[0040] As another objective of this invention, the present invention provides a fault and anomaly detection system for general aviation equipment, comprising:

[0041] Multi-source data acquisition and processing module: Through multiple sensors, CCTV system, network monitoring system and dispatch system, it collects the operation data of general aviation equipment in real time, and performs standardization processing, outlier removal and timestamp synchronization on the data to ensure data consistency and availability;

[0042] Dynamic Causal Relationship Network Construction Module: Based on the improved NOTEARS algorithm, a multi-layer dynamic causal network including the system macro layer and the device layer is constructed to generate a directed acyclic graph. The causal chain is optimized and learned through the gradient descent algorithm to analyze the potential nonlinear causal relationships between devices.

[0043] Real-time data stream processing module: Transmits real-time data through message queues and uses a stream processing engine to process real-time data streams. It performs feature extraction and preprocessing on device sensor data streams and log data streams to extract important time series features.

[0044] Fault detection and early warning module: Based on a multi-layer dynamic causal network, it analyzes real-time data to determine whether the current equipment status is abnormal, and infers the root cause of the fault through causal chain. When an abnormality is detected, it triggers an alarm and generates a fault report.

[0045] Compared with the prior art, the beneficial effects of the present invention include:

[0046] 1) This invention constructs a multi-layered dynamic causal relationship network comprising a system macro-level and a device level. The dynamic causal network at the system macro-level identifies the dependencies between various general aviation subsystems, while the dynamic causal network at the device level identifies potential fault chains between devices. By combining the distributed stream processing technology of the Apache Flink stream processing engine with the real-time message transmission mechanism of the Kafka message queue, it analyzes multi-source data from different general aviation devices in real time and extracts feature data, thereby achieving real-time automatic detection and location of equipment faults, improving fault handling efficiency, and effectively ensuring the safety and stability of general aviation.

[0047] 2) The improved NOTEARS algorithm, combined with a dynamic Bayesian network (DBN), divides the multi-source data of the general aviation system into time slices, introduces past observations as lagged variables, establishes causal relationships between lagged variables and current variables, facilitates the processing of data with varying time dimensions, and learns causal relationships that change over time. The graph structure of the existing NOTEARS algorithm is extended to a time-series structure, the static graph is improved to obtain a dynamic graph, and cross-time acyclic constraints and intra-time slice acyclic constraints are added, which facilitates the processing of nonlinear dynamic causal relationships of general aviation equipment, identifies potential fault chains between equipment, realizes causal chain reasoning of equipment faults, and helps to quickly locate the root cause of equipment fault anomalies.

[0048] 3) This invention effectively solves the complexity and latency problems of multi-source data analysis in general aviation scenarios by integrating multi-source data from general aviation system sensors, logs, network monitoring, etc., and using Kafka and Flink to achieve real-time data transmission and processing. It employs a multi-layered dynamic causal network model to achieve real-time detection of equipment anomalies and accurate reasoning of fault root causes, significantly reducing false alarms and missed alarms and improving the accuracy of fault location.

[0049] 4) The general aviation equipment fault and anomaly detection system of the present invention relies on multi-source data of the general aviation system and is based on a multi-layer dynamic causal relationship network to realize automated causal chain reasoning of equipment faults and anomalies and dynamic adjustment of detection indicators, thereby reducing the need for manual intervention and improving fault response speed and processing efficiency. Attached Figure Description

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] Figure 1 This is a schematic diagram of a general aviation equipment fault detection system according to an embodiment of the present invention.

[0052] Figure 2 This is a schematic diagram illustrating the process of processing multi-source data streams using a multi-layer dynamic causal relationship network, as described in an embodiment of the present invention. Detailed Implementation

[0053] like Figure 1As shown, the method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network includes:

[0054] Step S1: Collect multi-source data from the general aviation system;

[0055] The navigation equipment data includes real-time operating indicators of the equipment (such as temperature, pressure, speed, etc.), navigation equipment server logs (such as equipment start-up, shutdown, abnormal events, etc.), navigation monitoring data (such as sensor data, CCTV system, environmental information, etc.), navigation execution work order data (such as equipment ID, maintenance records, processing status, etc.), and navigation vessel information (such as vessel number, declaration time, anchoring time, etc.). This data is stored in a CSV file for easy reading by the algorithm.

[0056] Meanwhile, since the logs of the general aviation equipment server contain non-numerical data, they can be mapped to different numerical data for storage, as shown in Table 1.

[0057] Table 1 General Aviation Equipment Server Data

[0058]

[0059] The sample data for the indicators are shown in Table 2.

[0060] Table 2 Indicator Data

[0061]

[0062] Step S2: Preprocess the multi-source data, including data standardization, conversion into time series, and data timestamp synchronization;

[0063] This invention uses the z-score method to standardize the metrics of different devices. The z-score method formula is: where... This is the average value. The standard deviation is denoted as .

[0064] ;

[0065] For example, if a hydrological monitoring device detects a water level of 70m, the historical average water level is 65m, and the standard deviation is 5m, then the z-score is calculated as follows:

[0066] ;

[0067] Outlier data is removed using the three-standard-deviation method to ensure data accuracy. For example, water levels exceeding three standard deviations are considered outliers and removed before further processing.

[0068] Use the drain algorithm to extract navigation equipment server log data into events and convert them into time series. Use the TraceExtract algorithm to extract the critical path of the call chain data and convert it into a time series, which is convenient for subsequent dynamic causal network to learn the call relationship between services.

[0069] The TraceExtract algorithm takes the root span of the trace as input and calculates the critical path starting from its end. All child spans of the root span are sorted in descending order according to their end times.

[0070] The pseudocode is as follows:

[0071] 1 traceextract(root)

[0072] 2 if root.children is None or root.kind is Producer, then

[0073] 3 return [root]

[0074] 4 end if

[0075] 5 if root.kind is client, then

[0076] 6 return traceextract(root.chlid)

[0077] 7 end if

[0078] 8 path ← [root.end]

[0079] 9 children ← root.children sorted in descending order by end time

[0080] 10 lastChild ← children[0]

[0081] 11 path ← traceextract (lastChild).extend(path)

[0082] 12 for each span in children[1 :], do

[0083] 13 If span.endTime < lastChild.startTime, then

[0084] 14 path ← traceextract (span).extend(path)

[0085] 15lastChild ← span

[0086] 16End if

[0087] 17end for

[0088] 18path ← [root.start].extend(path)

[0089] 19 return path

[0090] 20end

[0091] At the same time, the timestamps of various data sources are aligned, enabling multi-dimensional time series analysis of general aviation equipment indicators, work orders, server logs, general aviation monitoring, and asset data to be performed synchronously.

[0092] Step S3: Using the improved NOTEARS algorithm, construct a multi-layered dynamic causal network, including constructing a dynamic causal network at the system macro level and a dynamic causal network at the equipment level. The dynamic causal network at the system macro level is used to identify the dependencies between various general aviation subsystems, and the dynamic causal network at the equipment level is used to handle the nonlinear dynamic causal relationships of general aviation equipment and identify potential fault chains between equipment, such as... Figure 2 As shown.

[0093] The improved NOTEARS algorithm improves upon the existing NOTEARS algorithm by including the following steps:

[0094] 1) Time series data modeling: The core of dynamic causal relationship modeling lies in processing data with changing time dimensions and learning the causal structure that changes over time on the time axis; dividing the time series data into time slices t1, t2...t T Each time slice corresponds to an observation over a specific period; past observations are introduced as lagged variables, such as X. t-3 X t-2 Possibly for the current X t The time period has a causal effect.

[0095] 2) Extend the graph structure of the NOTEARS algorithm to a temporal structure. In the original NOTEARS algorithm optimization problem, the dynamic causal relationship is a static graph. To convert it to a dynamic graph, the edges in the causal graph need to be changed to time-dependent edges, i.e., edges from time t−1 to time t. Accordingly, add time dimension constraints to ensure the acyclicity of the dynamic causal graph. The following cross-time acyclicity constraints are added to the original constraints:

[0096] ;

[0097] in This indicates the dynamic causal relationship of the current time slice. d represents the dynamic causal relationship of the previous time slice; ht() represents the number of variables in the causal network; Tr() represents the dynamic acyclic constraint function at time t; and Tr() represents the matrix trace operation function.

[0098] The Notears algorithm is used to learn the dynamic causal relationships within each time slice. Then, the framework of Dynamic Bayesian Network (DBN) is used to add directed edges between time slices, which represent the dependencies between lagged variables and current variables.

[0099] 3) Dynamic optimization: For each time slice, the original optimization strategy of Notears is still used, that is, minimizing the squared error or other loss function, to learn the dynamic causal relationship matrix W for each time slice. (t) The optimization objective is:

[0100] ;

[0101] in This is a matrix of lagged variables containing multiple time slices; The regularization term is denoted by ; loss() represents the loss function; ( ) represents a regularization function; This represents the regularization coefficient.

[0102] In this embodiment, step 3 specifically includes the following sub-steps:

[0103] Step S301: Feature selection and data transformation;

[0104] At the macro level of the general aviation system, based on business knowledge and data exploration results, time slices are divided according to the time dimension. Within each time slice, key features at the level of general aviation subsystems are selected (such as subsystem performance indicators, interdependent data traffic, etc.). The feature data of each general aviation subsystem (such as subsystem response time, error rate, data throughput, etc.) are represented as a feature matrix X. (t) The `_subsystems` field contains rows representing a point in time and columns representing characteristics of a subsystem. Example characteristics include the status of the reporting subsystem, the status of the navigation monitoring subsystem, vessel monitoring data, and lock operation status.

[0105] At the equipment level within the general aviation subsystem, for each general aviation subsystem, time slices are divided according to the time dimension. Within each time slice, relevant data from its internal equipment are selected, such as general aviation equipment sensor data, server operation logs, fault records, and internal server service call chains. A device-level feature matrix X is then constructed for each general aviation subsystem. (t)The `_devices` column represents a point in time in each row and a characteristic of a single device in each column. Example characteristics include: real-time device metrics such as temperature, pressure, network uplink and downlink speeds, packet loss rate, and memory pressure; work order data such as maintenance records and fault types; and server logs for general aviation equipment, including device anomaly events and device start / stop records.

[0106] Data transformation: Convert selected features into a feature matrix X (t) For example, server-related characteristics:

[0107] X (t) = ;

[0108] Step S302: Initialize the weighted adjacency matrix and Lagrange multipliers of the dynamic causal network, and construct the smoothing function.

[0109] At the macro level of the general aviation system, an initial weighted adjacency matrix W is selected for the optimization problem of the improved NOTEARS algorithm. (t) 0_subsystems and Lagrange multiplier α0; construct the smoothing function h(W (t) a) to enable it to encode acyclic constraints; construct ht(W (t) a) Ensure the acyclicity constraint of the dynamic causal graph in the time dimension;

[0110] At the device level within the general aviation subsystem, initialize the device-level weighted adjacency matrix W. (t) 0_devices and Lagrange multiplier β0; construct smoothing function h(W (t) b) to enable it to encode acyclic constraints; construct ht(W) (t) b) Ensure the acyclicity constraint of the dynamic causal graph in the time dimension;

[0111] Step S303: Optimize the weighted adjacency matrix to obtain a dynamic directed acyclic graph, i.e., a dynamic causal network.

[0112] Numerical optimization methods such as L-BFGS are used to minimize the NOTEARS objective function at both the macroscopic level and the equipment level within subsystems of the general aviation system, satisfying h(W (t) a)=0、h(W (t) b)=0 intra-time slice acyclic constraint and time-dimension acyclic constraint ht(W) (t) a)=0、ht(W (t) b)=0; simultaneously iteratively update W. (t) _devices and W (t) _subsystems, will optimize the matrix W (t) _devices and W (t)The subsystems are thresholded to obtain a dynamic directed acyclic graph (DAG), which serves as a dynamic causal network.

[0113] Step S304: Verify and correct the dynamic causal network.

[0114] At the macro level of the general aviation system, test data from general aviation subsystems are used to learn the dynamic causal network W between subsystems. (t) The _subsystems are used for verification. If an error chain is found, manual intervention is performed on the erroneous causal chain to correct it, and finally the dynamic causal network is determined.

[0115] At the device level within the general aviation subsystem, device-level test data is used to verify the dynamic causal network W between devices within each subsystem. (t) _devices, through manual intervention and correction of unreasonable dynamic causal relationship chains, ultimately confirm the dynamic causal network structure at the device level.

[0116] Given a W (t) Example: If there are three characteristic variables, namely, the network latency of the general aviation declaration system (latency), the bandwidth usage of the general aviation declaration system (bandwidth), and the packet loss rate of the general aviation declaration system (packet loss).

[0117] W (t) = ;

[0118] W (t) 12 =0.3 indicates that the network latency of the general aviation declaration system has a positive causal impact on the bandwidth usage of the general aviation declaration system.

[0119] W (t) 23 =0.4 indicates that the bandwidth usage of the general aviation declaration system has a positive causal impact on the packet loss rate of the general aviation declaration system.

[0120] W (t) 31 =−0.3 indicates that the packet loss rate of the general aviation declaration system has a negative causal impact on the network latency of the general aviation declaration system.

[0121] Directed Acyclic Graph (DAG): General Aviation Declaration System Delay → General Aviation Declaration System Bandwidth → General Aviation Declaration System Packet Loss Rate.

[0122] Step S4: Use a message queue to process the real-time data stream of the general aviation system and extract the feature data for real-time fault detection.

[0123] The various subsystems of the navigation system transmit real-time data via Kafka. For example, Kafka topics such as `network_metrics` and `device_metrics` are set up for transmitting network monitoring data and device metric data, respectively. As a message queue, Kafka ensures that real-time data collected from different data sources (such as the navigation network monitoring system and hydrological monitoring equipment sensors) is efficiently and timely transmitted to the processing system. Kafka producers send these real-time data streams from the data sources to the consumers, while Flink, as a Kafka consumer, subscribes to these topics to receive the real-time data streams.

[0124] Flink processes the received real-time data stream. It performs preprocessing, feature extraction, and real-time analysis of the data, conforming to step S2. z-score is used to standardize network latency and bandwidth usage data to ensure consistency and accuracy. Subsequently, key features, such as the maximum and average network latency and packet loss rate variations, are calculated using a sliding window technique, and these features are extracted for subsequent analysis.

[0125] Step S5: Use the dynamic causal relationship network obtained in step S3 to analyze the feature data obtained in step S4 to determine whether there is a device malfunction. If the determination result is yes, then further use the multi-layer dynamic causal relationship network to locate the cause of the device malfunction.

[0126] In this example, the system assesses whether the current system status is abnormal based on characteristic data such as network latency, bandwidth usage, and device metrics. If the characteristics in the real-time data stream exceed a predetermined threshold, the system will further analyze the data using a dynamic causal network model, generate a judgment result, and have it manually confirmed. If relevant changes occur, the threshold will be dynamically adjusted. This process includes calculating the device's health score, determining whether the device is abnormal based on the score, thereby triggering alarms and generating detailed fault reports. The device score is calculated using the following formula:

[0127] ;

[0128] Among them, W (t) Let X be a weighted adjacency matrix. (t) The characteristic matrix, This is the activation function.

[0129] Based on the Directed Acyclic Graph (DAG), trace the causal chain of anomalous variables to find possible upstream causes. For example, if the packet loss rate of the general aviation declaration system increases abnormally, the causal chain in the DAG will indicate that the packet loss rate is affected by network latency and bandwidth usage. Tracing back from the packet loss rate of the general aviation declaration system, examine the status of network latency and bandwidth usage to further identify potential causes of the anomaly. Combine this with actual data to determine the location of the fault and take further action.

[0130] Step S6: If there is a device malfunction based on the judgment result of step S5, a fault alarm will be issued.

[0131] The implementation results show that by constructing a multi-layered dynamic causal relationship network that includes a system macro-level and an equipment level, this invention achieves real-time and efficient detection and root cause localization of equipment faults, improves fault handling efficiency, and safeguards the safe operation of shipping systems.

[0132] Another embodiment of the present invention provides a fault detection system for general aviation equipment based on dynamic causal network analysis, such as... Figure 1 As shown, the system includes:

[0133] Multi-source data acquisition and processing module: This module collects operational data from multiple sensors and systems, including sensor data from navigation equipment such as hydrological monitors, navigation monitoring data (e.g., CCTV system data), navigation declaration system server data, network monitoring data, log data from the dispatching system, and manually entered data by staff. The data types are complex, including structured time-series data, unstructured log data, and inter-system call chain data. The module provides a given data input format and performs preprocessing accordingly. Specifically, log data can be extracted into events using the drain algorithm and sorted by time series; call chain data uses the TraceExtract algorithm to extract critical paths and sort them by time, facilitating subsequent dynamic causal network learning of service call relationships.

[0134] The dynamic multi-layer causal network construction module constructs a dynamic causal network between subsystems at the macro level, identifies the dependencies between subsystems, and then establishes a dynamic causal network for each device within each subsystem. An improved dynamic NOTEARS algorithm is used to construct the dynamic causal network. The improved dynamic NOTEARS algorithm on a single time slice transforms the causal structure learning problem into a differentiable optimization problem and ensures that the learned result is a directed acyclic graph through acyclic constraints. The graph network generated by the improved dynamic NOTEARS algorithm may contain errors; a dynamic feedback mechanism is introduced to improve the accuracy of the graph network construction. In the general aviation equipment management scenario, the dynamic causal network is used to analyze the causal chain of equipment failures. The input data of the improved dynamic NOTEARS algorithm are sensor data from the equipment, time-seriesified log data, and call chain data. The algorithm optimizes and learns dynamic causal relationships through gradient descent, thereby generating a dynamic causal relationship model between devices. The generated dynamic causal network not only reflects the correlation between devices but also allows for network structure adjustment through intervention operations, ultimately used for accurate fault source localization.

[0135] Real-time streaming data processing module: To perform real-time fault detection of general aviation equipment, the system transmits real-time data via Kafka, and Flink performs real-time processing and analysis of the data stream. In practical applications, the general aviation subsystem server transmits the collected equipment sensor data streams, log data streams, and call chain data streams via Kafka. Flink preprocesses these data streams, extracting important time-series features, and the anomaly detection server receives these data streams.

[0136] Fault Detection and Early Warning Module: The general aviation equipment anomaly detection server uses a dynamic causal network model to perform real-time fault detection of general aviation equipment and determine whether the current system status is abnormal. When an anomaly is detected, the dynamic causal network is used to deduce the root cause of the fault, and an alarm is triggered through the system. The results are then returned to the respective subsystem devices to prompt maintenance personnel to handle the issue promptly.

Claims

1. A method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal relationship network, characterized in that, Includes the following steps: Step S1: Collect multi-source data from the navigation system. The multi-source data includes navigation equipment data, navigation equipment server logs, navigation monitoring data, navigation execution work orders, and navigation vessel information data. Step S2: Preprocess the multi-source data, including data standardization, conversion into time series, and data timestamp synchronization; Step S3: Construct a multi-layer dynamic causal relationship network, including constructing a dynamic causal network at the system macro level and a dynamic causal network at the equipment level; the dynamic causal network at the system macro level is used to identify the dependencies between various general aviation subsystems, and the dynamic causal network at the equipment level is used to handle the nonlinear dynamic causal relationships of general aviation equipment and identify potential fault chains between equipment. Step S4: Use a message queue to process the real-time data stream of the general aviation system and extract the feature data for real-time fault detection. Step S5: Use the dynamic causal relationship network obtained in step S3 to analyze the feature data obtained in step S4 to determine whether there is a device malfunction. If the determination result is yes, then further use the dynamic causal relationship network to locate the cause of the device malfunction. Step S6: Based on the judgment result of step S5, if there is a device malfunction, a fault alarm will be issued; Step S3 uses an improved NOTEARS algorithm to construct a multi-layer dynamic causal network. The improvement of the existing NOTEARS algorithm by the improved NOTEARS algorithm includes the following steps: 1) Time series data modeling: The core of dynamic causal relationship modeling lies in processing data with changing time dimensions and learning the causal structure that changes over time on the time axis; dividing time series data into time slices, with each time slice corresponding to observations over a certain period of time; introducing past observations as lagged variables and establishing the causal relationship between lagged variables and current variables; 2) Extend the graph structure of the NOTEARS algorithm to a temporal structure. In the original optimization problem of the NOTEARS algorithm, the dynamic causal relationship is a static graph. Change the edges in the static graph to time-dependent edges to obtain a dynamic graph. Accordingly, add time dimension constraints to ensure the acyclicity of the dynamic causal graph. Add cross-time acyclicity constraints to the original NOTEARS algorithm constraints. The time-acyclic constraint is: ； in This indicates the dynamic causal relationship of the current time slice. d represents the dynamic causal relationship of the previous time slice; ht() represents the number of variables in the causal network; Tr() represents the dynamic acyclic constraint function at time t; Tr() represents the matrix trace operation function. The Notears algorithm is used to learn the dynamic causal relationships within each time slice, and then a dynamic Bayesian network is used to add directed edges between time slices. These directed edges are used to represent the dependencies between lagged variables and current variables. 3) Dynamic optimization: For each time slice, the original optimization strategy of Notears is still used, namely minimizing the squared error or other loss functions; The optimization objective for learning the dynamic causal relationship matrix for each time slice is: ； in This is a matrix of lagged variables containing multiple time slices; The regularization term is denoted by ; loss() represents the loss function; ( ) represents a regularization function; Represents the regularization coefficient; Step 3 specifically includes the following sub-steps: Step S301: Feature selection and data transformation; At the macro level of the general aviation system, based on business knowledge and data exploration results, time slices are divided according to the time dimension, and key features at the level of general aviation subsystems are selected within each time slice; the feature data of each general aviation subsystem is represented as a feature matrix X. (t) _subsystems; At the equipment level within the general aviation subsystem, for each general aviation subsystem, time slices are divided according to the time dimension, and relevant data of its internal equipment are selected within each time slice; a feature matrix X at the equipment level is constructed for each general aviation subsystem. (t) _devices; Step S302: Initialize the weighted adjacency matrix and Lagrange multipliers of the dynamic causal network, and construct the smoothing function; At the macro level of the general aviation system, an initial weighted adjacency matrix W is selected for the optimization problem of the improved NOTEARS algorithm. (t) 0_subsystems and Lagrange multiplier α0; constructing the system-level intra-time slice smoothing function h(W (t) a) Enables it to encode acyclic constraints; constructs acyclic constraint function ht(W) between system-level time slices. (t) a) is used to ensure the acyclicity constraint of dynamic causal graphs from the time dimension; At the device level within the general aviation subsystem, initialize the device-level weighted adjacency matrix W. (t) 0_devices and Lagrange multiplier β0; construct the device-level time-slice smoothing function h(W (t) b) Enables it to encode acyclic constraints; constructs acyclic constraint function ht(W) between device-level time slices. (t) b) is used to ensure the acyclicity constraint of the dynamic causal graph from the time dimension; Step S303: Optimize the weighted adjacency matrix to obtain a dynamic directed acyclic graph, i.e., a dynamic causal network; Numerical optimization methods are used to minimize the NOTEARS objective function at both the macroscopic level and the equipment level within subsystems of the general aviation system, satisfying h(W (t) a)=0、h(W (t) b)=0 intra-time slice acyclic constraint and time-dimension acyclic constraint ht(W) (t) a)=0、ht(W (t) b)=0; simultaneously iteratively update the system-level weighted adjacency matrix W. (t) _devices and the device-level weighted adjacency matrix W (t) _subsystems, will optimize the matrix W (t) _devices and W (t) After thresholding, the subsystems are transformed into a dynamic directed acyclic graph, which serves as a dynamic causal network. Step S304: Verify and correct the dynamic causal network; At the macro level of the general aviation system, test data from the general aviation subsystems are used to verify the learned dynamic causal network between the general aviation subsystems. If an incorrect causal chain is found, manual intervention and correction are carried out on the incorrect causal chain to finally determine the system-level dynamic causal network. At the equipment level within the general aviation subsystem, equipment-level test data is used to verify the dynamic causal network between devices within each general aviation subsystem. Through manual intervention and correction of unreasonable dynamic causal relationship chains, the dynamic causal network structure at the equipment level is finally determined.

2. The method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal network according to claim 1, characterized in that, Step S2 specifically includes the following sub-steps: S201: Use the z-score method to standardize the indicator data of different devices; S202: Use the three-standard-deviation method to remove outliers and ensure data accuracy; S203: Use the drain algorithm to extract general aviation equipment server log data into events and convert it into time series, extract the key path of call chain data and convert it into time series; S204: Timestamp synchronization of data for each category enables multivariate time series analysis of general aviation equipment indicators, work orders, server logs, general aviation monitoring and asset data.

3. The method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal network according to claim 1 or 2, characterized in that, In step S4, the real-time data stream of the aviation system is transmitted using the message queue Kafka, and the real-time data stream is processed using the stream processing engine Flink. The feature data in the real-time data stream is analyzed and extracted using a sliding window.

4. The method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal network according to claim 3, characterized in that, In step S5, based on the multi-layer dynamic causal relationship network obtained in step S3, the feature data of the real-time data stream provided by the Flink streaming engine is analyzed. The health score of the device is calculated according to the feature data and compared with a predetermined threshold to determine whether the device status is abnormal. The causal chain of the device abnormality is traced through the directed acyclic graph of the multi-layer dynamic causal relationship network to find the root cause of the device abnormality.

5. The method for detecting faults and anomalies in general aviation equipment based on a multi-layer dynamic causal network according to claim 4, characterized in that, Step S5 calculates the device's health score using the following formula: ； in, W represents the device health score. (t) Let X be a weighted adjacency matrix. (t) The characteristic matrix, This is the activation function.

Images