A fault assessment and localization system for aviation communications

Abstract

The application relates to the technical field of aviation communication, and discloses a fault evaluation and positioning system for aviation communication, which comprises a communication link data acquisition module, an anomaly detection and fault classification module, a multi-source data fusion fault positioning module, a fault risk prediction and hierarchical early warning module and a fault repair and network optimization module. Through real-time data acquisition, anomaly detection and multi-source data fusion, the application can quickly identify and accurately position aviation communication link faults, utilize time difference positioning and Kalman filter optimization results, combine fault risk prediction and hierarchical early warning, provide accurate repair suggestions and network optimization schemes, and improve the stability and operation and maintenance efficiency of the communication link.

Description

Technical Field

[0001] This application relates to the technical field of aviation communications, and in particular to a fault assessment and location system for aviation communications. Background Technology

[0002] Aviation communications are communication systems between aircraft and ground control, other aircraft, or airborne equipment, encompassing various forms such as air-to-ground, air-to-air, and airborne communication. The main communication technologies include VHF radio, HF radio, and satellite communication (SATCOM). These technologies ensure flight safety, route management, emergency response, and weather information sharing. Data link communications (such as ACARS) provide real-time flight plans, route adjustments, and flight status updates.

[0003] With the advancement of aviation communication technology, challenges include signal attenuation, interference, and spectrum management; future development trends include the integration of 5G technology and AI to promote low-latency, high-speed communication, enhance automated and intelligent monitoring and prediction capabilities, thereby improving flight safety and efficiency.

[0004] Existing technologies have certain limitations in fault mode identification and classification, resulting in insufficient accuracy and timeliness in fault diagnosis. At the same time, existing fault location technologies may face problems such as low location accuracy or noise interference. In addition, traditional technologies usually rely solely on real-time data for fault processing, lacking the ability to predict and anticipate potential faults. Summary of the Invention

[0005] To address the issues of low positioning accuracy or noise interference that may exist in existing fault location technologies, this application provides a fault assessment and location system for aviation communications.

[0006] This application provides a fault assessment and location system for aviation communications, employing the following technical solution:

[0007] A fault assessment and location system for aviation communications, the fault assessment and location system comprising:

[0008] The communication link data acquisition module collects key performance indicators of the communication link through multiple monitoring points, including airborne equipment deployed in the air and ground stations on the ground.

[0009] The anomaly detection and fault classification module performs real-time anomaly detection on key performance indicators based on the operation and maintenance perception classification model, diagnoses and classifies the detected anomalies, and determines the specific fault type.

[0010] The multi-source data fusion fault location module matches the corresponding location algorithm based on the fault type and integrates real-time data from multiple monitoring points. By analyzing key observations of signal transmission, it can locate the fault source in the spatial or logical link dimension and determine the specific location where the fault occurred.

[0011] The fault risk prediction and graded early warning module analyzes long-term historical data and fault types to predict the probability of fault risk for a specific link or device within a preset time period and issue graded early warnings in advance.

[0012] The fault repair and network optimization module automatically generates differentiated repair suggestions based on the fault type and specific location, and formulates long-term network optimization plans based on the fault risk probability.

[0013] Optionally, the anomaly detection and fault classification module includes:

[0014] The feature engineering and dimensionality reduction module extracts discriminative time-domain and frequency-domain features based on the collected key performance indicators of the communication link, and uses mutually exclusive feature binding technology to fuse and reduce the dimensionality of the time-domain and frequency-domain features to obtain a low-dimensional feature set.

[0015] The fault classification model training module uses the operation and maintenance perception classification model to train the feature set. By minimizing the loss function and maximizing the information gain of each split, it gradually builds multiple weak classification trees and learns the feature distribution and classification boundary under different fault modes.

[0016] The real-time anomaly detection and identification module inputs the low-dimensional feature set generated in real time into the trained operation and maintenance perception classification model, outputs the predicted probability of each performance indicator belonging to various types of faults, and identifies abnormal performance indicators by comparing the predicted probability with a preset threshold.

[0017] The fault diagnosis and classification module determines the fault category based on the abnormal performance indicators and classifies the fault category into a preset fault type.

[0018] Optionally, the feature set is trained using an operation and maintenance awareness classification model. By minimizing the loss function and maximizing the information gain of each split, multiple weak classification trees are gradually constructed, and the feature distribution and classification boundaries under different fault modes are learned, including:

[0019] Obtain a feature set containing performance metrics from the dataset, define a loss function as the optimization objective, and initialize the model's predictions.

[0020] In each iteration, for each performance metric, the first and second derivatives of the corresponding loss function are calculated;

[0021] Based on the first and second derivatives, all split points are traversed, the information gain of each split point is calculated, and the split point with the largest information gain is selected as the splitting basis for the current round of weak classification tree.

[0022] The weak classification tree for the current round is constructed based on the selected split point, and the output of each weak classification tree is multiplied by the learning rate and then weighted and accumulated into the prediction value of the current model.

[0023] Repeat the process of calculating gradients, selecting split points, building and updating weak classification trees until the set termination condition is met. The final operation and maintenance awareness classification model is obtained by weighted sum of the prediction results of all weak classification trees.

[0024] Optionally, the expression for calculating the information gain at each split point is:

[0025]

[0026] In the formula, This represents the information gain at the current split point; Represents the feature set of the left node; i Indicates performance metrics; Represents the feature set of the left node; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the left child node; Represents the regularization parameter; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the right child node; It represents the first derivative of the loss function with respect to the predicted value.

[0027] Optionally, the expression for the predicted probability of each performance metric belonging to each type of fault is as follows:

[0028]

[0029] In the formula, Indicates a given input performance metric At that time, performance indicators Category The probability of; Indicates category; Index representing the category; express b The index value of the category score; This indicates the total number of categories into which the performance metric is divided.

[0030] Optionally, the multi-source data fusion fault location module includes:

[0031] The multi-source data synchronization and correlation module collects real-time communication signal data from airborne equipment and ground monitoring points, and ensures data time synchronization to obtain a multi-source real-time dataset associated with known fault types.

[0032] The preliminary location module selects a matching time difference location algorithm based on the known fault type and calculates the signal arrival time difference based on the multi-source real-time dataset to obtain the preliminary location results of the fault type for each monitoring point.

[0033] The data fusion and regional refinement module fuses the preliminary positioning results of each monitoring point and uses Kalman filtering to eliminate single-point data errors and noise. Combined with the spatial geometric features or logical link rules corresponding to the fault type, it narrows down the candidate fault area and locks the final fault location range.

[0034] The multi-feature aggregation and locking module associates all key observation features within the final fault location range and accurately locks the fault source to a specific physical device or logical node through multi-feature aggregation analysis.

[0035] Optionally, based on the known fault type, a matching time difference localization algorithm is selected, and the signal arrival time difference is calculated based on a multi-source real-time dataset to obtain preliminary fault type localization results for each monitoring point, including:

[0036] Based on the identified fault type, the time difference positioning algorithm is called from the preset strategy library, and the corresponding list of valid monitoring points and calibration parameters are loaded.

[0037] Identify associated signals caused by the same fault source from multi-source data, extract the signal arrival time of each monitoring point, and calibrate all timestamps to a unified precision time reference;

[0038] Select a monitoring point that meets the preset requirements as a reference point, calculate the signal arrival time difference between other monitoring points and the reference point, and construct a corresponding positioning hyperboloid based on each time difference;

[0039] By solving the spatial intersection region of multiple positioning hyperboloids, the geometric center of the spatial intersection region is output as the preliminary spatial coordinates of the fault source;

[0040] The expression for constructing the corresponding hyperboloid based on each time difference is as follows:

[0041]

[0042]

[0043]

[0044] In the formula, Indicates the distance from the point to be located to the first...i The straight-line distance between the monitoring points; Indicates the distance from the point to be located to the reference point. j The straight-line distance; Indicates the speed at which the signal propagates; Indicates the signal has arrived at the first i Each monitoring point and the reference point j Time difference; x , y , z () represents the coordinates of the signal source in three-dimensional space; x i , y i , z i ) indicates the first i The three-dimensional coordinates of each monitoring point (non-reference point); x j , y j , z j () indicates the reference point j The three-dimensional coordinates.

[0045] Optionally, the preliminary location results of each monitoring point are fused, and Kalman filtering is used to eliminate single-point data errors and noise. Combined with the spatial geometric features or logical link rules corresponding to the fault type, the candidate fault area is narrowed down, and the final fault location range is determined, including:

[0046] Receive asynchronous preliminary positioning results from each monitoring point, and perform coordinate transformation and time alignment operations on them according to a unified spatiotemporal reference to form standardized multi-source data;

[0047] The standardized multi-source data is input into the Kalman filter, and the fault source state is estimated iteratively through prediction and update steps to output the preliminary fault source location.

[0048] The initial fault source location is mapped to the actual spatial geometry or network topology. Unreasonable candidate points are eliminated by using link connectivity rules or spatial constraints, narrowing down the location area and obtaining an optimized set of candidate points.

[0049] Based on the optimized candidate point set, the weighted centroid and boundary range are calculated, and the final fault location range with geographic coordinates or logical identifiers is output.

[0050] Optionally, the expression for calculating the weighted centroid is:

[0051]

[0052] In the formula, Indicates the weighted centroid; This represents the total number of candidate points in the optimized candidate point set; Indicate candidate points; Indicates the first n The weights of each candidate point; Indicates the first n Candidate points at x Coordinate values ​​on the axis; Indicates the first n Candidate points at y Coordinate values ​​on the axis; Indicates the first n Candidate points at z coordinates on the axis

[0053] Optionally, the fault risk prediction and graded early warning module includes:

[0054] The historical sample construction module integrates long-term historical performance data and confirmed fault records, extracts temporal features that characterize the health status of equipment, and labels the data with fault status within a preset future time period to construct a historical sample set for model training.

[0055] The risk probability prediction module, based on historical sample sets, divides the data into training and test sets, uses a time-series machine learning model for training, and outputs a quantified probability of failure risk.

[0056] The early warning strategy management module sets high, medium, and low thresholds for the fault risk probability output by the model based on operation and maintenance costs, forming a hierarchical early warning trigger rule base.

[0057] The real-time early warning execution module compares the calculated failure risk probability with the hierarchical early warning trigger rule base. If the probability exceeds the critical threshold, the corresponding level of early warning will be automatically triggered.

[0058] In summary, this application includes at least one of the following beneficial technical effects:

[0059] 1. This application rapidly identifies and accurately locates aviation communication link faults through real-time data acquisition, anomaly detection, and multi-source data fusion. It utilizes time difference positioning, Kalman filtering optimization results, and combines fault risk prediction and graded early warning to provide accurate repair suggestions and network optimization solutions, thereby improving the stability and operation and maintenance efficiency of the communication link.

[0060] 2. This application optimizes the split point by maximizing information gain, making the classification boundary more closely match the actual fault mode; at the same time, it achieves refined diagnosis by predicting the probability output of various faults, which helps to realize rapid identification, accurate classification and proactive handling of faults, and significantly improves the intelligent operation and maintenance level and fault response efficiency of aviation communications.

[0061] 3. This application significantly improves the accuracy and reliability of fault location by combining real-time data from airborne equipment and ground monitoring points and employing technologies such as time difference positioning, Kalman filtering, and multi-feature aggregation. Through time synchronization and data fusion, noise is eliminated and the candidate fault area is narrowed down, ultimately pinpointing the location of the fault source. Combined with fault risk prediction and graded early warning based on historical data, potential fault risks can be anticipated in advance and early warnings can be automatically triggered, thereby improving the stability and response efficiency of aviation communications. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the system in this application. Detailed Implementation

[0063] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0064] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0065] This application discloses a fault assessment and location system for aviation communications, referring to... Figure 1 The fault assessment and location system includes:

[0066] The communication link data acquisition module collects key performance indicators of the communication link through multiple monitoring points, including airborne equipment deployed in the air and ground stations on the ground.

[0067] It should be explained that airborne equipment usually refers to communication equipment deployed on aircraft (such as drones, satellites, airplanes, etc.), which is responsible for communicating with ground stations. They usually carry sensors, data acquisition equipment and communication hardware to monitor and transmit data in real time. Ground stations are facilities deployed on the ground to establish communication links with airborne equipment. Ground stations usually have the ability to receive, process and store data.

[0068] To comprehensively monitor the health status of communication links, multiple monitoring points will be set up in the air (airborne equipment) and on the ground (ground stations). These monitoring points may include radio wave signal receivers, data transmission interfaces, signal quality detection instruments, etc.

[0069] Key performance indicators (KPIs) are standards for evaluating the quality and health of communication links. Common KPIs for communication links include:

[0070] 1. Signal strength: The signal strength of the communication link affects the reliability of the link. A weak signal may lead to packet loss or unstable connection.

[0071] 2. Signal-to-noise ratio (SNR): The higher the ratio of signal to noise, the better the communication quality; a low SNR indicates poor link quality.

[0072] 3. Latency: refers to the time required for data to be transmitted from the sending end to the receiving end. A longer latency may affect the performance of real-time applications.

[0073] 4. Packet loss rate: The proportion of data lost during transmission. A high packet loss rate indicates instability in the communication link.

[0074] 5. Bandwidth utilization: The bandwidth usage of the communication link. Excessive bandwidth utilization may lead to network congestion.

[0075] 6. Bit error rate: The ratio of errors that occur during transmission. A high bit error rate may lead to poor data quality.

[0076] The anomaly detection and fault classification module performs real-time anomaly detection on key performance indicators based on the operation and maintenance perception classification model, diagnoses and classifies the detected anomalies, and determines the specific fault type.

[0077] Preferably, the anomaly detection and fault classification module includes:

[0078] The feature engineering and dimensionality reduction module extracts discriminative time-domain and frequency-domain features based on the collected key performance indicators of the communication link, and uses mutually exclusive feature binding technology to fuse and reduce the dimensionality of the time-domain and frequency-domain features to obtain a low-dimensional feature set.

[0079] It should be explained that time-domain characteristics refer to the performance of a signal over time; for example: average, maximum, minimum, fluctuation range, standard deviation, skewness, kurtosis, autocorrelation, and cross-correlation.

[0080] Frequency domain features are those extracted by converting a signal from the time domain to the frequency domain through spectral analysis; examples include spectral density, dominant frequency, bandwidth, and frequency response characteristics of the signal.

[0081] The purpose of mutual exclusion feature binding technology is to effectively fuse time-domain features and frequency-domain features, reduce redundant features, and ensure that they can work together to provide more discriminative information. The specific implementation is as follows:

[0082] 1. Feature Fusion: Time-domain features and frequency-domain features represent different dimensions of a signal, describing the signal's properties from the perspectives of time and frequency, respectively. By fusing these features, the signal's characteristics can be captured more comprehensively, improving the performance of subsequent models.

[0083] 2. Mutually exclusive feature binding: This process ensures that the features extracted from the time domain and frequency domain can complement each other and avoid redundancy. For example, some time domain features may already contain information from frequency domain features. Mutually exclusive binding technology will identify and remove these redundancies while retaining the most discriminative information.

[0084] By combining mutually exclusive feature binding techniques with dimensionality reduction, a low-dimensional feature set is obtained. This set can retain key discriminative information while removing redundant features. The low-dimensional feature set is the foundation for training the model and can help improve the model's efficiency and performance.

[0085] The fault classification model training module uses the operation and maintenance awareness classification model to train the feature set. By minimizing the loss function and maximizing the information gain of each split, it gradually builds multiple weak classification trees and learns the feature distribution and classification boundary under different fault modes.

[0086] Preferably, the specific steps for training the feature set using an operation and maintenance awareness classification model, gradually constructing multiple weak classification trees by minimizing the loss function and maximizing the information gain of each split, and learning the feature distribution and classification boundary under different fault modes are as follows:

[0087] Obtain a feature set containing performance metrics from the dataset, define a loss function as the optimization objective, and initialize the model's predictions.

[0088] In each iteration, for each performance metric, the first and second derivatives of the corresponding loss function are calculated;

[0089] Based on the first and second derivatives, all split points are traversed, the information gain of each split point is calculated, and the split point with the largest information gain is selected as the splitting basis for the current round of weak classification tree.

[0090] The weak classification tree for the current round is constructed based on the selected split point, and the output of each weak classification tree is multiplied by the learning rate and then weighted and accumulated into the prediction value of the current model.

[0091] Repeat the process of calculating gradients, selecting split points, building and updating weak classification trees until the set termination condition is met. The final operation and maintenance awareness classification model is obtained by weighted sum of the prediction results of all weak classification trees.

[0092] Preferably, the expression for calculating the information gain at each split point is:

[0093]

[0094] In the formula, This represents the information gain at the current split point; Represents the feature set of the left node; i Indicates performance metrics; Represents the feature set of the left node; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the left child node; Represents the regularization parameter; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the right child node; It represents the first derivative of the loss function with respect to the predicted value.

[0095] The real-time anomaly detection and identification module inputs the low-dimensional feature set generated in real time into the trained operation and maintenance perception classification model, outputs the predicted probability of each performance indicator belonging to various types of faults, and identifies abnormal performance indicators by comparing the predicted probability with a preset threshold.

[0096] Preferably, the expression for the predicted probability of each performance index belonging to each type of fault is as follows:

[0097]

[0098] In the formula, Indicates a given input performance metric At that time, performance indicators Category The probability of; Indicates category; Index representing the category; express b The index value of the category score; This indicates the total number of categories into which the performance metric is divided.

[0099] The fault diagnosis and classification module determines the fault category based on the abnormal performance indicators and classifies the fault category into a preset fault type.

[0100] Specific examples are illustrated below:

[0101] In large-scale UAV monitoring systems, UAVs and ground control centers interact via wireless communication links. To ensure communication stability and data security during mission execution, real-time health monitoring and fault diagnosis of the communication links are required.

[0102] Deployment method: The UAV (airborne end) is equipped with a communication transmission and data acquisition module, and multiple monitoring stations are deployed on the ground (ground end);

[0103] Data collection metrics: Signal strength index (RSSI), signal-to-noise ratio (SNR), data transmission delay, packet loss rate, bandwidth utilization, and bit error rate;

[0104] For each communication link metric (such as RSSI, SNR, etc.), the following time-domain features are extracted: maximum value, minimum value, average value, variance, skewness, autocorrelation coefficient, etc.

[0105] Given the signal strength (RSSI) data of 5 samples: 85, 78, 92, 80, 90, the following time-domain characteristics can be calculated:

[0106] Maximum value: 92;

[0107] Minimum value: 78;

[0108] Average: 36;

[0109] Skewness: 0.23;

[0110] Autocorrelation coefficient: 0.89.

[0111] Applying Fast Fourier Transform (FFT) to the RSSI signal yields the spectrum, and then frequency domain features can be extracted.

[0112] Suppose we perform an FFT on 5 sample values ​​of the signal strength to obtain the following spectral data:

[0113] Spectral density FFT (RSSI) = [5, 3, 7, 2, 6];

[0114] The dominant frequency is the frequency position where the maximum value is located;

[0115] The frequency bandwidth is the distance from the first non-zero frequency to the last non-zero frequency;

[0116] The spectral density is the amplitude at each frequency point.

[0117] Combine time-domain features (such as maximum, minimum, and average values) with frequency-domain features (such as dominant frequency, bandwidth, and spectral density) into a new feature set;

[0118] The fused feature vector: fused feature = [maximum RSSI value, minimum RSSI value, average RSSI value, RSSI spectral density, ...];

[0119] By calculating the correlation between features, highly correlated features can be removed; for example, if the maximum and minimum values ​​of RSSI are highly correlated, only the maximum value can be retained.

[0120] Principal component analysis (PCA) is used to reduce the dimensionality of the fused features, retaining the most discriminative features and reducing the dimensionality of the data.

[0121] Assuming we use principal component analysis for dimensionality reduction, compressing the 5 features into 2 principal components, we obtain the following low-dimensional features:

[0122] Table 1 Low-dimensional feature table

[0123]

[0124] Low-dimensional features will be used as input for subsequent fault classification models;

[0125] Ultimately, the feature vector of each sample will include a fusion of time-domain and frequency-domain features. After redundancy elimination and dimensionality reduction, a low-dimensional feature set is obtained; for example:

[0126] Table 2 Low-dimensional feature set table

[0127]

[0128] Low-dimensional features are input into the classification model to predict different types of faults (such as channel interference, link congestion, etc.).

[0129] We have already calculated a split point (signal strength 75dBm). Now, we will continue to show how to explore other potential split points, such as splitting based on bandwidth utilization.

[0130] Assume that bandwidth utilization is chosen as a characteristic, and it is decided to split at 70%;

[0131] Step 1: Select a bandwidth utilization rate of 70% as the split point;

[0132] Left child node (L): contains samples with bandwidth utilization of less than or equal to 70%;

[0133] Samples 1, 2, and 4 (bandwidth utilization: 60%, 70%, and 80%)

[0134] Right child node (R): contains samples with bandwidth utilization greater than 70%;

[0135] Samples 3 and 5 (bandwidth utilization: 55% and 65%)

[0136] Using the previously calculated first derivative ( g i The first derivative is the same as before:

[0137] g 1 = -0.2 g 2 = -1.1 g 3 = 0.85 g 4 = -0.3 and g 5 = 0.95;

[0138] Calculate the second derivative (H L and H R );

[0139] left child node ( L ):

[0140] The gradient values ​​for samples 1, 2, and 4 are: g 1 = -0.2, g 2 = -1.1, g 4 = -0.3;

[0141] Calculate the second derivative of the left child node (i.e., the variance of the sample): H L =1.34;

[0142] The gradient values ​​for samples 3 and 5 are: g 3 = 0.85 g 5 = 0.95;

[0143] Calculate the second derivative of the right child node: H R =1.625;

[0144] Calculate the gradient sum of the left and right child nodes. =-1.6; =1.8;

[0145] The final information gain is =3.35;

[0146] Based on the above information gain calculation, it can be seen that the information gain corresponding to a signal strength of 75dBm and a bandwidth utilization of 70% is 3.65. Since the information gain is the same, other features and split points can be further evaluated.

[0147] Next, the new tree is used to update the residuals of the samples (i.e., the difference between the predicted and the true values), and this is gradually improved in subsequent iterations.

[0148] During training, the model builds multiple weak classification trees through multiple iterations. In each iteration, the error is gradually reduced based on the gradient and information gain. After training all the trees, the model integrates these trees by weighting to obtain the final strong classifier and completes the training of the model.

[0149] Each time a new data sample (e.g., signal strength, signal-to-noise ratio, bandwidth utilization, etc.) is received, the data is input into the previously trained model, which calculates the predicted probability for each fault category; for example:

[0150] Probability of interference failure: 0.78;

[0151] Probability of congestion failure: 0.15;

[0152] Probability of normal state: 0.07;

[0153] Set a threshold (e.g., 0.7). If the probability of a certain fault category is greater than this threshold, the sample is considered to be abnormal.

[0154] Obtain anomaly information from the real-time anomaly detection module;

[0155] Based on preset rules (e.g., signal strength below 80dBm is considered an interference fault), the specific type of fault is further confirmed. If the signal strength is detected to be below 80dBm and other indicators also meet the characteristics of an interference fault, it is judged as an "interference fault," the specific fault type is determined, and it is classified.

[0156] The multi-source data fusion fault location module matches the corresponding location algorithm based on the fault type and integrates real-time data from multiple monitoring points. By analyzing key observations of signal transmission, it can pinpoint the fault source in the spatial or logical link dimension and determine the specific location where the fault occurred.

[0157] Preferably, the multi-source data fusion fault location module includes:

[0158] The multi-source data synchronization and correlation module collects real-time communication signal data from airborne equipment and ground monitoring points, and ensures data time synchronization to obtain a multi-source real-time dataset associated with known fault types.

[0159] The preliminary location module selects a matching time difference location algorithm based on the known fault type and calculates the signal arrival time difference based on the multi-source real-time dataset to obtain the preliminary location results of the fault type for each monitoring point.

[0160] Preferably, the specific implementation steps for selecting a matching time difference localization algorithm based on the known fault type and calculating the signal arrival time difference based on a multi-source real-time dataset to obtain the preliminary fault type localization results for each monitoring point are as follows:

[0161] Based on the identified fault type, the time difference positioning algorithm is called from the preset strategy library, and the corresponding list of valid monitoring points and calibration parameters are loaded.

[0162] Identify associated signals caused by the same fault source from multi-source data, extract the signal arrival time of each monitoring point, and calibrate all timestamps to a unified precision time reference;

[0163] Select a monitoring point that meets the preset requirements as a reference point, calculate the signal arrival time difference between other monitoring points and the reference point, and construct a corresponding positioning hyperboloid based on each time difference;

[0164] The expression for constructing the corresponding hyperboloid based on each time difference is as follows:

[0165]

[0166]

[0167]

[0168] In the formula, Indicates the distance from the point to be located to the first... i The straight-line distance between the monitoring points; Indicates the distance from the point to be located to the reference point. j The straight-line distance; Indicates the speed at which the signal propagates; Indicates the signal has arrived at the first i Each monitoring point and the reference point j Time difference; x , y , z () represents the coordinates of the signal source in three-dimensional space; x i , y i , z i ) indicates the first i The three-dimensional coordinates of each monitoring point (non-reference point); x j , y j , z j () indicates the reference point j The three-dimensional coordinates.

[0169] By solving the spatial intersection region of multiple positioning hyperboloids, the geometric center of the spatial intersection region is output as the preliminary spatial coordinates of the fault source.

[0170] Uniform sampling is performed within the intersection region enclosed by the three hyperboloids. n 1 effective point ( n It needs to be large enough to ensure the entire coverage area is covered, and the coordinates of the sampling points should be remembered:

[0171] ( x 1, y 1, z 1), x 2, y 2, z 2), ..., ( x n , y n , zn );

[0172] right n sampling points x , y , z The coordinates are taken as the arithmetic mean, and the result is the preliminary spatial coordinates of the fault source. x 0, y 0, z 0), the expression is as follows:

[0173]

[0174] In the formula, ( x 0, y 0, z 0) represents the initial spatial coordinates of the fault source;

[0175] , , Indicates the first within the intersection area k The coordinates of a discrete sampling point;

[0176] n This represents the total number of discrete sampling points (arbitrarily set). n The larger the initial coordinates, the closer they are to the actual center of the region.

[0177] k This represents discrete sampling points.

[0178] It should be explained that, among all the monitoring points, one monitoring point that meets the preset accuracy and geometric distribution requirements is selected as the reference point. Then, the signal arrival time difference of each of the remaining monitoring points relative to the reference point is calculated. Each time difference corresponds to a set of possible fault source locations, and the fault source locations form a hyperboloid in three-dimensional space. Subsequently, all hyperboloids are solved jointly, that is, their spatial intersection region is calculated. This region represents the possible location of the fault source when multiple monitoring points observe the same time difference. By calculating the geometric center of this intersection region, the preliminary spatial coordinates of the fault source are obtained.

[0179] The data fusion and regional refinement module fuses the preliminary positioning results of each monitoring point and uses Kalman filtering to eliminate single-point data errors and noise. Combined with the spatial geometric features or logical link rules corresponding to the fault type, it narrows down the candidate fault area and locks the final fault location range.

[0180] Preferably, the preliminary location results of each monitoring point are fused, and Kalman filtering is used to eliminate single-point data errors and noise. Combined with the spatial geometric features or logical link rules corresponding to the fault type, the candidate fault area is narrowed down, and the final fault location range is locked. The specific implementation steps are as follows:

[0181] Receive asynchronous preliminary positioning results from each monitoring point, and perform coordinate transformation and time alignment operations on them according to a unified spatiotemporal reference to form standardized multi-source data;

[0182] The standardized multi-source data is input into the Kalman filter, and the fault source state is estimated iteratively through prediction and update steps to output the preliminary fault source location.

[0183] The initial fault source location is mapped to the actual spatial geometry or network topology. Unreasonable candidate points are eliminated by using link connectivity rules or spatial constraints, thus narrowing down the location area and obtaining an optimized set of candidate points.

[0184] It needs to be explained that the preliminary positioning results (such as 3D coordinates or logical node numbers) are projected onto the actual Geographic Information System (GIS) or network topology. For example, the positioning points are associated with known devices, links, or nodes through coordinate matching or topology mapping. Then, based on the link connectivity (such as whether there is a physical or logical connection between devices) or spatial rules (such as whether the area is reachable or not deployed), candidate points are screened to exclude those that are disconnected from the backbone link, are in blind spots, or violate geographical restrictions (such as coordinates falling outside the wall or outside the monitoring range). This eliminates unreasonable points and ultimately retains points within the range of effective links and in areas allowed by spatial rules, forming a smaller and more reliable optimized candidate set.

[0185] Based on the optimized candidate point set, the weighted centroid and boundary range are calculated, and the final fault location range with geographic coordinates or logical identifiers is output.

[0186] The expression for calculating the weighted centroid is:

[0187]

[0188] In the formula, Indicates the weighted centroid; This represents the total number of candidate points in the optimized candidate point set; Indicate candidate points; Indicates the first n The weights of each candidate point; Indicates the first n Candidate points at x Coordinate values ​​on the axis; Indicates the first n Candidate points aty Coordinate values ​​on the axis; Indicates the first n Candidate points at z The coordinate values ​​on the axis.

[0189] Find the minimum and maximum values ​​of all candidate points in the x, y, and z dimensions respectively: Based on the minimum and maximum values ​​in the x, y, and z dimensions, obtain the spatial range of the fault location, that is, the boundary range. Through this clear boundary range, finally output the fault location range with geographic coordinates or logical identifiers.

[0190] The multi-feature aggregation and locking module associates all key observation features within the final fault location range and accurately locks the fault source to a specific physical device or logical node through multi-feature aggregation analysis.

[0191] The fault risk prediction and graded early warning module analyzes long-term historical data and fault types to predict the probability of fault risk for a specific link or device within a preset time period and issues graded early warnings in advance.

[0192] Preferably, the fault risk prediction and graded early warning module includes:

[0193] The historical sample construction module integrates long-term historical performance data and confirmed fault records, extracts temporal features that characterize the health status of equipment, and labels the data with fault status within a preset future time period to construct a historical sample set for model training.

[0194] The risk probability prediction module, based on historical sample sets, divides the data into training and test sets, uses a time-series machine learning model for training, and outputs a quantified probability of failure risk.

[0195] The early warning strategy management module sets high, medium, and low thresholds for the fault risk probability output by the model based on operation and maintenance costs, forming a hierarchical early warning trigger rule base.

[0196] The real-time early warning execution module compares the calculated failure risk probability with the hierarchical early warning trigger rule base. If the probability exceeds the critical threshold, the corresponding level of early warning will be automatically triggered.

[0197] The fault repair and network optimization module automatically generates differentiated repair suggestions based on the fault type and specific location, and formulates long-term network optimization plans based on the fault risk probability.

[0198] It should be noted that high risk indicates a high probability of failure, which may lead to serious consequences;

[0199] Medium risk indicates a moderate probability of failure, which may affect performance but will not immediately lead to serious problems;

[0200] Low risk means that the probability of failure is low and can be ignored or routine monitoring can be performed.

[0201] By acquiring the status data of each device or network node in real time, the current probability of failure risk is predicted; at this time, the risk prediction module outputs the current failure risk value based on the latest data.

[0202] The risk probability calculated in real time is compared with the warning threshold in the rule base, for example:

[0203] If the calculated risk probability is greater than the high-risk threshold, a high-risk warning will be automatically triggered, and relevant personnel will be notified to take action.

[0204] If the risk probability is in the middle range, a medium-risk warning will be triggered, and some preventive measures will be recommended.

[0205] If the risk probability is low, only routine checks may be performed and logs may be recorded.

[0206] Long-term network optimization solutions include: in high-risk areas, it is recommended to add redundant equipment, increase bandwidth, or rearrange the network topology to distribute the load; in medium- and low-risk areas, performance optimization measures may be taken to improve resource utilization efficiency or adjust load balancing.

[0207] Specific examples are as follows:

[0208] After confirming that it was an interference fault, the multi-source data synchronization and correlation module collected data from different monitoring points, including real-time signal data from ground monitoring points and airborne equipment. Through precise time synchronization, it ensured that the data timestamps were consistent, and all data were correlated with the type of interference fault, forming a multi-source dataset.

[0209] After the fault type is confirmed, the preliminary location module selects the time difference of arrival (TDOA) positioning algorithm and loads the relevant monitoring point list and calibration parameters according to the preset strategy library. At this time, by comparing the arrival time difference of the interference source signals received by different monitoring points, the approximate location of the fault source is calculated using the TDOA positioning algorithm. For example, the preliminary location result shows that the fault source is located in a 100m × 100m area.

[0210] At this point, the data fusion and regional refinement module takes over the positioning process. It optimizes the preliminary positioning results through Kalman filtering technology, performs coordinate transformation and time alignment on the positioning data from various monitoring points, and forms a standardized multi-source data input Kalman filter. The filter iteratively calculates to eliminate noise and single-point data errors, and obtains a more accurate fault source location. By combining spatial constraints and network topology rules, unreasonable positioning candidate points are eliminated, and finally the fault source is located in a more accurate area, and the location of the fault source is determined to be (50, 50) meters.

[0211] Next, we analyzed the key characteristics of all monitoring points in the area, such as signal strength and bit error rate. Through aggregation analysis, we identified the association between the fault source and a certain base station device, and finally pinpointed the fault source to the device of that base station.

[0212] After determining the location and type of the fault, the fault risk prediction and graded early warning module begins to function. It analyzes the historical performance data of the base station equipment and uses a time series machine learning model to predict the probability of the equipment failing in the future. The model results show that the fault risk of the equipment has a 50% probability of exceeding the preset high-risk threshold in the next 24 hours, thus triggering a high-risk early warning and notifying the operation and maintenance personnel to conduct an inspection.

[0213] Finally, the fault repair and network optimization module generated repair suggestions based on the fault type and location. It suggested adjusting the frequency of the base station equipment to avoid interference, and also suggested replacing the equipment or adjusting the signal coverage to optimize the communication quality of the area. In addition, based on the fault risk prediction results, it proposed a long-term network optimization plan, which includes adding backup equipment in the area, optimizing the network topology, and implementing a regular frequency adjustment strategy.

[0214] It should be noted that the calculation formulas and all parameters involved in the calculations in this application have been dimensionless beforehand. The process of dimensionless processing is well known in the industry and will not be described here.

[0215] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A fault assessment and location system for aviation communications, characterized in that, The fault assessment and location system includes: The communication link data acquisition module collects key performance indicators of the communication link through multiple monitoring points, including airborne equipment deployed in the air and ground stations on the ground. The anomaly detection and fault classification module performs real-time anomaly detection on key performance indicators based on the operation and maintenance perception classification model, diagnoses and classifies the detected anomalies, and determines the specific fault type. The anomaly detection and fault classification module includes: The feature engineering and dimensionality reduction module extracts discriminative time-domain and frequency-domain features based on the collected key performance indicators of the communication link, and uses mutually exclusive feature binding technology to fuse and reduce the dimensionality of the time-domain and frequency-domain features to obtain a low-dimensional feature set. The fault classification model training module uses the operation and maintenance perception classification model to train the feature set. By minimizing the loss function and maximizing the information gain of each split, it gradually builds multiple weak classification trees and learns the feature distribution and classification boundary under different fault modes. The real-time anomaly detection and identification module inputs the low-dimensional feature set generated in real time into the trained operation and maintenance perception classification model, outputs the predicted probability of each performance indicator belonging to various types of faults, and identifies abnormal performance indicators by comparing the predicted probability with a preset threshold. The process of training the feature set using an operation and maintenance awareness classification model involves gradually constructing multiple weak classification trees by minimizing the loss function and maximizing the information gain of each split, and learning the feature distribution and classification boundaries under different fault modes, including: Obtain a feature set containing performance metrics from the dataset, define a loss function as the optimization objective, and initialize the model's predictions. In each iteration, for each performance metric, the first and second derivatives of the corresponding loss function are calculated; Based on the first and second derivatives, all split points are traversed, the information gain of each split point is calculated, and the split point with the largest information gain is selected as the splitting basis for the current round of weak classification tree. The weak classification tree for the current round is constructed based on the selected split point, and the output of each weak classification tree is multiplied by the learning rate and then weighted and accumulated into the prediction value of the current model. Repeat the process of calculating gradients, selecting split points, building and updating weak classification trees until the set termination conditions are met. The final operation and maintenance awareness classification model is obtained by weighted sum of the prediction results of all weak classification trees. The fault diagnosis and classification module determines the fault category based on the abnormal performance indicators and classifies the fault category into a preset fault type; The multi-source data fusion fault location module matches the corresponding location algorithm based on the fault type and fuses real-time data from multiple monitoring points. By analyzing key observations of signal transmission, it locks the fault source in the spatial or logical link dimension and determines the specific location where the fault occurred. The fault risk prediction and graded early warning module analyzes long-term historical data and fault types to predict the probability of fault risk for a specific link or device within a preset time period and issue graded early warnings in advance. The fault repair and network optimization module automatically generates differentiated repair suggestions based on the fault type and specific location, and formulates long-term network optimization plans based on the fault risk probability.

2. The fault assessment and location system for aviation communications according to claim 1, characterized in that, The expression for calculating the information gain at each split point is: ； In the formula, This represents the information gain at the current split point; Represents the feature set of the left node; i Indicates performance metrics; Represents the feature set of the left node; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the left child node; Represents the regularization parameter; This represents the sum of the second derivatives of the loss function with respect to the predicted values ​​for all performance metrics in the right child node; It represents the first derivative of the loss function with respect to the predicted value.

3. The fault assessment and location system for aviation communications according to claim 2, characterized in that, The expression for predicting the probability of each performance index belonging to a certain type of fault is as follows: ； In the formula, Indicates the performance index of a given input At that time, performance indicators Category The probability of; Indicates category; Index representing the category; express b The index value of the category score; This indicates the total number of categories into which the performance metric is divided.

4. The fault assessment and location system for aviation communications according to claim 3, characterized in that, The multi-source data fusion fault location module includes: The multi-source data synchronization and correlation module collects real-time communication signal data from airborne equipment and ground monitoring points, and ensures data time synchronization to obtain a multi-source real-time dataset associated with known fault types. The preliminary location module selects a matching time difference location algorithm based on the known fault type and calculates the signal arrival time difference based on the multi-source real-time dataset to obtain the preliminary location results of the fault type for each monitoring point. The data fusion and regional refinement module fuses the preliminary positioning results of each monitoring point and uses Kalman filtering to eliminate single-point data errors and noise. Combined with the spatial geometric features or logical link rules corresponding to the fault type, it narrows down the candidate fault area and locks the final fault location range. The multi-feature aggregation and locking module associates all key observation features within the final fault location range and accurately locks the fault source to a specific physical device or logical node through multi-feature aggregation analysis.

5. The fault assessment and location system for aviation communications according to claim 4, characterized in that, The process of selecting a matching time difference localization algorithm based on the known fault type and calculating the signal arrival time difference based on a multi-source real-time dataset to obtain preliminary fault type localization results for each monitoring point includes: Based on the identified fault type, the time difference positioning algorithm is called from the preset strategy library, and the corresponding list of valid monitoring points and calibration parameters are loaded. Identify associated signals caused by the same fault source from multi-source data, extract the signal arrival time of each monitoring point, and calibrate all timestamps to a unified precision time reference; Select a monitoring point that meets the preset requirements as a reference point, calculate the signal arrival time difference between other monitoring points and the reference point, and construct a corresponding positioning hyperboloid based on each time difference; By solving the spatial intersection region of multiple positioning hyperboloids, the geometric center of the spatial intersection region is output as the preliminary spatial coordinates of the fault source; The expression for constructing the corresponding hyperboloid based on each time difference is as follows: ； ； ； In the formula, Indicates the distance from the point to be located to the first... i The straight-line distance between the monitoring points; Indicates the distance from the point to be located to the reference point. j The straight-line distance; Indicates the speed at which the signal propagates; Indicates the signal has arrived at the first i Each monitoring point and the reference point j Time difference; x , y , z () represents the coordinates of the signal source in three-dimensional space; x i , y i , z i ) indicates the first i The three-dimensional coordinates of each monitoring point; x j , y j , z j () indicates the reference point j The three-dimensional coordinates.

6. The fault assessment and location system for aviation communications according to claim 5, characterized in that, The process of fusing the preliminary location results from each monitoring point, using Kalman filtering to eliminate single-point data errors and noise, and combining the spatial geometric features or logical link rules corresponding to the fault type to narrow down the candidate fault area and pinpoint the final fault location range includes: Receive asynchronous preliminary positioning results from each monitoring point, and perform coordinate transformation and time alignment operations on them according to a unified spatiotemporal reference to form standardized multi-source data; The standardized multi-source data is input into the Kalman filter, and the fault source state is estimated iteratively through prediction and update steps to output the preliminary fault source location. The initial fault source location is mapped to the actual spatial geometry or network topology. Unreasonable candidate points are eliminated by using link connectivity rules or spatial constraints, narrowing down the location area and obtaining an optimized set of candidate points. Based on the optimized candidate point set, the weighted centroid and boundary range are calculated, and the final fault location range with geographic coordinates or logical identifiers is output.

7. The fault assessment and location system for aviation communications according to claim 6, characterized in that, The expression for calculating the weighted centroid is: ； In the formula, Indicates the weighted centroid; This represents the total number of candidate points in the optimized candidate point set; Indicate candidate points; Indicates the first n The weights of each candidate point; Indicates the first n Candidate points at x Coordinate values ​​on the axis; Indicates the first n Candidate points at y Coordinate values ​​on the axis; Indicates the first n Candidate points at z The coordinate values ​​on the axis.

8. The fault assessment and location system for aviation communications according to claim 1, characterized in that, The fault risk prediction and graded early warning module includes: The historical sample construction module integrates long-term historical performance data and confirmed fault records, extracts temporal features that characterize the health status of equipment, and labels the data with fault status within a preset future time period to construct a historical sample set for model training. The risk probability prediction module, based on historical sample sets, divides the data into training and test sets, uses a time-series machine learning model for training, and outputs a quantified probability of failure risk. The early warning strategy management module sets high, medium, and low thresholds for the fault risk probability output by the model based on operation and maintenance costs, forming a hierarchical early warning trigger rule base. The real-time early warning execution module compares the calculated failure risk probability with the hierarchical early warning trigger rule base. If the probability exceeds the critical threshold, the corresponding level of early warning will be automatically triggered.

Images