Navigation hardware fault diagnosis system based on adaptive clustering algorithm

Abstract

The invention discloses a navigation hardware fault diagnosis system based on an adaptive clustering algorithm, relates to the field of satellite navigation systems, and solves the technical problems that multi-dimensional feature extraction and comprehensive evaluation value calculation are difficult to carry out, and a new fault type is difficult to verify by setting multiple judgment results. Comprising the following steps: collecting multi-source data from a satellite, a user receiver and a ground control system; key features are extracted from each type of processed data, and signal, positioning and hardware circuit comprehensive evaluation values are calculated; feature vectors are constructed, after standardization, Eps parameters of the DBSCAN are dynamically determined through the K distance, and self-adaptive clustering is achieved; based on a clustering result, defining a similarity calculation rule, constructing a known fault type library, identifying fault types through a matching rule, and analyzing clusters which cannot be matched; and if the clustering cannot be matched and meets various judgment conditions, identifying the fault type as a new fault type, updating the characteristics of the fault type to a known fault library, and triggering model retraining at the same time.

Description

Technical Field

[0001] The present invention belongs to the field of satellite navigation systems, and in particular relates to a navigation hardware fault diagnosis system based on an adaptive clustering algorithm. Background Art

[0002] Satellite navigation systems play a vital role in many fields such as modern communications, transportation, and military. However, due to their complex structure and the special environment in which they are located, hardware failures often occur in satellite navigation systems.

[0003] Traditional fault diagnosis methods often rely on manual experience and simple testing tools. For complex and diverse hardware fault types, the diagnostic efficiency is low and the accuracy is difficult to guarantee. Especially when facing various possible hardware faults such as signal reception failure, user receiver failure and hardware circuit failure, it is difficult to dynamically adjust parameters to improve clustering stability, and it is also difficult to combine the known fault type library with the new fault identification mechanism. The lack of a fast, accurate and adaptive fault diagnosis system can easily lead to long troubleshooting time, affecting the normal operation of the navigation system and the continuity of service.

[0004] Therefore, a navigation hardware fault diagnosis system based on adaptive clustering algorithm came into being. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a navigation hardware fault diagnosis system based on an adaptive clustering algorithm to solve the following technical problems:

[0006] Traditional fault diagnosis methods often rely on manual experience and simple testing tools. For complex and diverse hardware fault types, the diagnostic efficiency is low and the accuracy is difficult to guarantee. Especially when facing various possible hardware faults such as signal reception failure, user receiver failure and hardware circuit failure, it is difficult to dynamically adjust parameters to improve clustering stability, and it is also difficult to combine the known fault type library with the new fault identification mechanism. The lack of a fast, accurate and adaptive fault diagnosis system can easily lead to long troubleshooting time, affecting the normal operation of the navigation system and the continuity of service.

[0007] To solve the above problems, the present invention provides a navigation hardware fault diagnosis system based on an adaptive clustering algorithm, which includes the following modules:

[0008] Data acquisition module: collects operating status data from the satellite, obtains the quality data of the received signal and its own status data from the user receiver, and obtains monitoring data on the satellite's operating status from the ground control system;

[0009] Data processing and feature extraction module: pre-processes the collected data, extracts key features from each type of processed data, and calculates the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value respectively;

[0010] Adaptive algorithm clustering module: Constructs a three-dimensional feature vector from the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value, performs normalization, calculates the K distance of each sample, determines the EPS parameter through the ascending sorting graph, inputs the normalized feature vector and parameters, performs cluster analysis, and outputs the cluster label;

[0011] Fault diagnosis and classification module: Based on the results of the adaptive clustering algorithm, similarity calculation and matching rules are defined, a library of known fault types is built to match the clustering results, and clusters that cannot be directly matched to known fault types are analyzed;

[0012] New fault type identification and fault library update module: If the clustering cannot be matched, and the dynamic threshold, fusion similarity and multi-judgment results do not match, it is identified as a new fault type and updated to the known fault type library.

[0013] Preferably, the data acquisition module includes:

[0014] The operating status data includes various operating parameters of the satellite equipment: signal strength, frequency stability and attitude information;

[0015] The quality data and status data include signal-to-noise ratio, bit error rate, and receiving sensitivity; the status information of the user receiver itself, including device temperature, battery level, and processor load;

[0016] The monitoring data includes satellite orbit parameters, ground station communication link status and detection records of various satellite functions.

[0017] Preferably, extracting key features from each type of processed data includes:

[0018] Establish a data processing server in the ground data processing center to receive data from all sources, clean the data, remove noise using a data filtering algorithm, and normalize the denoised data;

[0019] Extracting key features from the processed data, including signal reception fault features, positioning error fault features, and hardware circuit fault features;

[0020] Among them, the signal reception fault characteristics include signal strength related characteristics, signal-to-noise ratio characteristics and bit error characteristics; the positioning error fault characteristics include position deviation characteristics, position stability characteristics and satellite geometric distribution related characteristics; the hardware circuit fault characteristics include voltage and current characteristics, temperature characteristics and circuit characteristics related to signal quality.

[0021] Preferably, the comprehensive signal evaluation value includes:

[0022] Count the relevant characteristics of the received signal strength within a preset time period, including the mean, variance and change rate, and take the average value to obtain the first signal evaluation value;

[0023] Obtain the signal-to-noise ratio data of the signal, calculate the mean and standard deviation of the signal-to-noise ratio to obtain the signal-to-noise ratio variation coefficient as the second evaluation value of the signal;

[0024] At the same time, the error characteristics of the received data are counted, including the error rate, continuous error length, error distribution unevenness and error interval, and the error characteristics are weighted and added to obtain the third evaluation value of the signal;

[0025] The error distribution unevenness is obtained by collecting the error information received by the navigation system, classifying the error information according to actual needs, and calculating the probability of each error information by counting the number of occurrences. Specifically,

[0026]

[0027] Among them, R B is the unevenness of error distribution, p i is the probability of the i-th type of error, and n is the number of types of error;

[0028] According to the importance of signal strength related features, signal-to-noise ratio features and error features to signal reception fault features, weight coefficients are assigned and weighted addition is performed to obtain the comprehensive signal evaluation value.

[0029] Preferably, the comprehensive positioning evaluation value includes:

[0030] Calculate the average deviation between the user receiver position and the true position, as well as the degree of fluctuation of the deviation; calculate the rate of change of the position over time for longitude, latitude, and altitude respectively, and take the average; detect whether the position jumps, and count the number of jumps to obtain the position jump frequency; count the number of currently visible satellites and their position information, and calculate the geometric precision factor;

[0031] Each feature is standardized; the standardized features are weighted and combined to obtain a comprehensive positioning evaluation value.

[0032] Preferably, the comprehensive evaluation value of the hardware circuit includes:

[0033] The voltage and current characteristics include voltage mean, current mean, voltage standard deviation, and current standard deviation; the temperature characteristics include temperature mean and temperature standard deviation; the signal quality-related characteristics include radio frequency signal characteristics and intermediate frequency signal characteristics, wherein the radio frequency signal characteristics include radio frequency signal strength and signal-to-noise ratio, and the intermediate frequency signal characteristics include intermediate frequency signal amplitude and distortion;

[0034] The features are normalized, and the normalized feature values ​​are weighted and added to obtain a comprehensive evaluation value of the hardware circuit.

[0035] Preferably, the adaptive algorithm clustering module includes:

[0036] The comprehensive evaluation value of signal reception, comprehensive evaluation value of positioning and comprehensive evaluation value of hardware circuit are used as feature vectors, and each sample corresponds to a three-dimensional feature vector;

[0037] The three comprehensive evaluation values ​​are standardized;

[0038] Calculate the distance from each sample point to its nearest neighbor, that is, K distance, where K distance is equal to min_samples; draw an ascending sort graph of K distances and find the inflection point. The distance corresponding to the inflection point is eps;

[0039] Input the standardized feature vector, parameters min_samples and eps, use Python's sklearn.cluster.DBSCAN for clustering, and output the cluster label;

[0040] The cluster labels: -1 represents a noise point that cannot be classified into any cluster; 0 to n represent different cluster labels, and each cluster represents a potential fault type.

[0041] Preferably, the fault diagnosis and classification module includes:

[0042] Calculate the characteristic center of each cluster and analyze its characteristic value distribution;

[0043] Build a library of known fault types, including fault type names, feature vectors, and similarity thresholds;

[0044] The similarity calculation is specifically as follows:

[0045]

[0046] Among them, S is the similarity, x j is the eigenvalue of the cluster center, y j is the characteristic value of a known fault type;

[0047] Define matching rules: If the similarity is less than the similarity threshold, the cluster is identified as a known fault type; otherwise, the cluster is identified as an unmatched cluster and analyzed.

[0048] Preferably, the identifying the cluster as an unmatched cluster and analyzing it includes:

[0049] Define the basic threshold as the global feature mean, calculate the current cluster density and global data density, and jointly construct the dynamic threshold. Specifically:

[0050]

[0051] Among them, T D is the dynamic threshold, T B is the basic threshold, D C is the current cluster density, D G is the global data density;

[0052] The KL divergence, time series cross similarity and Euclidean distance of the comprehensive feature distribution are used to assign weight coefficients and construct the fusion similarity, which is as follows:

[0053] S W =K d ×W1+S tc ×W2+D e ×W3

[0054] Among them, S W To fusion similarity, K d is the KL divergence, S tc is the temporal cross similarity, D e is the Euclidean distance, W1, W2 and W3 are the corresponding weight coefficients;

[0055] Calculate the sample feature statistics and draw a histogram to observe the feature distribution pattern; compare the cluster features with the features of the known fault model; calculate the Pearson correlation coefficient between the features; and make the first judgment: whether it is a new fault type;

[0056] Perform trend decomposition, sliding window detection, and time series pattern matching on time series data to determine whether it is a known fault; perform a second judgment: whether it is a time series fault or a new fault type;

[0057] Combine temperature, humidity, and pressure environmental data to perform correlation analysis, threshold detection, and joint clustering; and make a third judgment: whether it is an environment-related fault or a new fault type.

[0058] Preferably, the new fault type identification and fault library update module includes:

[0059] Define identification rules: If the cluster cannot match a known fault type and meets the following conditions, it is represented as a new fault type;

[0060] The following conditions include that neither the dynamic threshold nor the fusion similarity calculation matches a known fault type, and the first, second, and third judgment results are all new fault types;

[0061] The new fault type information includes the fault type name, feature vector, similarity threshold and additional information, wherein the additional information includes time series pattern and environmental sensitivity description; the new fault type is added to the known fault type library;

[0062] At the same time, each time a new fault type is identified, the library of known fault types is automatically updated and the model is retrained.

[0063] Beneficial effects of the present invention:

[0064] The present invention comprehensively reflects the system status and improves the accuracy of fault diagnosis by combining satellite operating status, user receiver signal quality, and ground control system monitoring data. It realizes the construction of a comprehensive evaluation value of multi-dimensional data, avoids the limitations of a single data source, calculates the K distance of each sample, and draws an ascending sorting graph to find the inflection point. It dynamically determines the EPS parameter of DBSCAN, reduces manual intervention, and improves the robustness of the clustering results. It also sets a dynamic threshold based on the known fault type distribution, avoiding the limitations of a fixed threshold. It integrates multiple similarity calculation methods to improve matching accuracy, and combines multiple judgment results for verification to ensure the reliability of new fault types. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of the module flow of the present invention;

[0066] Figure 2 Schematic diagram of the method flow of the present invention;

[0067] Figure 3 Schematic diagram of the multi-judgment result verification process of the present invention. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0069] See also Figure 1 As shown, the present invention is a navigation hardware fault diagnosis system based on an adaptive clustering algorithm, comprising the following modules:

[0070] Data acquisition module: collects operating status data from the satellite, obtains the quality data of the received signal and its own status data from the user receiver, and obtains monitoring data on the satellite's operating status from the ground control system;

[0071] Data processing and feature extraction module: pre-processes the collected data, extracts key features from each type of processed data, and calculates the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value respectively;

[0072] Adaptive algorithm clustering module: Constructs a three-dimensional feature vector from the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value, performs normalization, calculates the K distance of each sample, determines the EPS parameter through the ascending sorting graph, inputs the normalized feature vector and parameters, performs cluster analysis, and outputs the cluster label;

[0073] Fault diagnosis and classification module: Based on the results of the adaptive clustering algorithm, similarity calculation and matching rules are defined, a library of known fault types is built to match the clustering results, and clusters that cannot be directly matched to known fault types are analyzed;

[0074] New fault type identification and fault library update module: If the clustering cannot be matched, and the dynamic threshold, fusion similarity and multi-judgment results do not match, it is identified as a new fault type and updated to the known fault type library.

[0075] Specifically, multi-source data is collected from satellites, user receivers and ground control systems, the collected data is preprocessed and the comprehensive evaluation values ​​of signal reception, positioning and hardware circuits are extracted, a three-dimensional feature vector is constructed, the feature vector is standardized, the K distance is calculated to determine the EPS parameter, and DBSCAN is used for clustering. The clustering results are matched with the known fault type library, and the unmatched clusters are marked as unknown fault types. New fault types are identified through dynamic thresholds, fusion similarity, and multi-judgment result verification, the known fault type library is updated, and the model is retrained.

[0076] In one embodiment of the present invention, the data acquisition module includes:

[0077] The operating status data includes various operating parameters of the satellite equipment: signal strength, frequency stability and attitude information;

[0078] The quality data and status data include signal-to-noise ratio, bit error rate, and receiving sensitivity; the status information of the user receiver itself, including device temperature, battery level, and processor load;

[0079] The monitoring data includes satellite orbit parameters, ground station communication link status and detection records of various satellite functions.

[0080] In one embodiment of the present invention, extracting key features from each type of processed data includes:

[0081] Establish a data processing server in the ground data processing center to receive data from all sources, clean the data, remove noise using a data filtering algorithm, and normalize the denoised data;

[0082] Extracting key features from the processed data, including signal reception fault features, positioning error fault features, and hardware circuit fault features;

[0083] Among them, the signal reception fault characteristics include signal strength related characteristics, signal-to-noise ratio characteristics and bit error characteristics; the positioning error fault characteristics include position deviation characteristics, position stability characteristics and satellite geometric distribution related characteristics; the hardware circuit fault characteristics include voltage and current characteristics, temperature characteristics and circuit characteristics related to signal quality.

[0084] Specifically, data is received from multiple channels such as satellites, receivers and ground control systems, and data from different sources are converted into a unified format. Data cleaning includes missing value processing, outlier detection and duplicate data removal. Data filtering and denoising include selecting filtering algorithms and configuring filtering parameters, and dynamically adjusting filtering parameters according to data characteristics, such as the covariance matrix of Kalman filtering; calculating the global statistical parameters of each feature, and applying the normalization formula to convert the entire data to ensure that the dimensions of different features are consistent.

[0085] In one embodiment of the present invention, the comprehensive signal evaluation value includes:

[0086] Count the relevant characteristics of the received signal strength within a preset time period, including the mean, variance and change rate, and take the average value to obtain the first signal evaluation value;

[0087] Obtain the signal-to-noise ratio data of the signal, calculate the mean and standard deviation of the signal-to-noise ratio to obtain the signal-to-noise ratio variation coefficient as the second evaluation value of the signal;

[0088] At the same time, the error characteristics of the received data are counted, including the error rate, continuous error length, error distribution unevenness and error interval, and the error characteristics are weighted and added to obtain the third evaluation value of the signal;

[0089] The error distribution unevenness is obtained by collecting the error information received by the navigation system, classifying the error information according to actual needs, and calculating the probability of each error information by counting the number of occurrences. Specifically,

[0090]

[0091] Among them, R B is the unevenness of error distribution, pi is the probability of the i-th type of error, and n is the number of types of error;

[0092] According to the importance of signal strength related features, signal-to-noise ratio features and error features to signal reception fault features, weight coefficients are assigned and weighted addition is performed to obtain the comprehensive signal evaluation value.

[0093] Specifically, the length of the preset time period is selected based on the application scenario and signal characteristics. For example, for real-time fault detection, the preset time period is selected to be a few seconds to a few minutes; for signal fluctuation analysis and fault diagnosis, the preset time period is selected to be tens of minutes to a few hours; the received signal strength data within the preset time period is collected, and the following characteristics are calculated: mean, variance, and rate of change, and the average is taken to obtain the first signal evaluation value; the signal-to-noise ratio data within the preset time period is collected, and the mean and standard deviation of the signal-to-noise ratio are calculated, and the coefficient of variation of the signal-to-noise ratio is calculated as the second signal evaluation value, specifically: Among them, S2 is the second evaluation value of the signal, CV is the coefficient of variation of the signal-to-noise ratio, SNR std is the standard deviation, SNR avg The bit error data within a preset time period is collected, and the following features are calculated, including the bit error rate, the length of continuous bit errors, and the unevenness of the bit error distribution. Weight coefficients are assigned according to the importance of the bit error features, and weighted addition is performed to obtain the third evaluation value of the signal. According to the importance of the signal strength-related features, the signal-to-noise ratio features, and the bit error features to the signal reception fault features, weight coefficients are assigned and the comprehensive evaluation value of the signal is calculated, specifically:

[0094] S=α1*S1+α2*S2+α3*S3

[0095] Among them, S is the comprehensive evaluation value of the signal, S1, S2 and S3 are the first, second and third evaluation values ​​of the signal respectively, α1, α2 and α3 are the corresponding weight coefficients, and the values ​​are 0.5, 0.2 and 0.3 respectively.

[0096] In one embodiment of the present invention, the comprehensive positioning evaluation value includes:

[0097] Calculate the average deviation between the user receiver position and the true position, as well as the degree of fluctuation of the deviation; calculate the rate of change of the position over time for longitude, latitude, and altitude respectively, and take the average; detect whether the position jumps, and count the number of jumps to obtain the position jump frequency; count the number of currently visible satellites and their position information, and calculate the geometric precision factor;

[0098] Each feature is standardized; the standardized features are weighted and combined to obtain a comprehensive positioning evaluation value.

[0099] Specifically, the user receiver's position estimate data and actual position data are collected to ensure that the two sets of data are time-aligned. The deviations are calculated for longitude, latitude, and altitude respectively. The longitude, latitude, and altitude deviations are summed and divided by 3 to obtain the average deviation. The standard deviation is calculated as the degree of deviation fluctuation. The change rate is calculated for longitude, latitude, and altitude respectively, and the average is taken to obtain the position change rate evaluation value. The jump threshold is set according to actual needs, and the position data is traversed to compare whether the position change of adjacent data points exceeds the threshold. If the threshold is exceeded, it is recorded as a jump, and the jump frequency is counted. Specifically, Among them, P is the hopping frequency, N t is the number of jumps, N p is the total number of data points; collect the number of currently visible satellites and their location information, including the satellite's azimuth and elevation angles, and calculate the GDOP using the geometric dilution of precision formula; use the Z-score normalization method to standardize each feature, where the mean and standard deviation are calculated based on the historical data set; assign weight coefficients based on the importance of each feature, and perform weighted addition to calculate the comprehensive positioning evaluation value, specifically:

[0100] G=β1*G1+β2*G2+β3*G3+P*β4+GDOP*β5

[0101] Among them, G is the comprehensive evaluation value of positioning, G1 is the mean deviation, G2 is the standard deviation, that is, the degree of fluctuation, G3 is the evaluation value of the position change rate, P is the hopping frequency, GDOP is the geometric precision dilution, β1, β2, β3, β4 and β5 are the corresponding weight coefficients, with values ​​of 0.2, 0.2, 0.3, 0.15 and 0.15 respectively.

[0102] In one embodiment of the present invention, the comprehensive evaluation value of the hardware circuit includes:

[0103] The voltage and current characteristics include voltage mean, current mean, voltage standard deviation, and current standard deviation; the temperature characteristics include temperature mean and temperature standard deviation; the signal quality-related characteristics include radio frequency signal characteristics and intermediate frequency signal characteristics, wherein the radio frequency signal characteristics include radio frequency signal strength and signal-to-noise ratio, and the intermediate frequency signal characteristics include intermediate frequency signal amplitude and distortion;

[0104] The features are normalized, and the normalized feature values ​​are weighted and added to obtain a comprehensive evaluation value of the hardware circuit.

[0105] Specifically, the voltage, current, temperature and signal quality related data of the hardware circuit are collected, wherein the signal quality related data include RF signal data and intermediate frequency signal data, and the collected data are cleaned, including removing outliers and filling missing values; the features of the processed data are extracted separately, and the current and voltage features include the voltage mean, current mean, voltage standard deviation and current standard deviation, the temperature features include the temperature mean and temperature standard deviation, the RF signal features include the RF signal strength and signal-to-noise ratio, and the intermediate frequency signal features include the intermediate frequency signal amplitude and distortion; the mean and standard deviation of each feature are calculated respectively, and each feature is standardized using the Z-score normalization method; weight coefficients are assigned according to the importance of each feature, and weighted addition is performed to obtain a comprehensive evaluation value of the hardware circuit, wherein the weight coefficient corresponding to each feature value is 0.1.

[0106] In one embodiment of the present invention, the adaptive algorithm clustering module includes:

[0107] The comprehensive evaluation value of signal reception, comprehensive evaluation value of positioning and comprehensive evaluation value of hardware circuit are used as feature vectors, and each sample corresponds to a three-dimensional feature vector;

[0108] The three comprehensive evaluation values ​​are standardized;

[0109] Calculate the distance from each sample point to its nearest neighbor, that is, K distance, where K distance is equal to min_samples; draw an ascending sort graph of K distances and find the inflection point. The distance corresponding to the inflection point is eps;

[0110] Input the standardized feature vector, parameters min_samples and eps, use Python's sklearn.cluster.DBSCAN for clustering, and output the cluster label;

[0111] The cluster labels: -1 represents a noise point that cannot be classified into any cluster; 0 to n represent different cluster labels, and each cluster represents a potential fault type.

[0112] Specifically, the comprehensive evaluation value of signal reception, the comprehensive evaluation value of positioning, and the comprehensive evaluation value of hardware circuit are used as feature vectors. Each sample corresponds to a three-dimensional feature vector, and the three comprehensive evaluation values ​​are standardized to eliminate the dimension effect. The K distance of each sample point is calculated. The K distance refers to the maximum distance from each sample point to its min_samples nearest neighbor points. The K distance is calculated using sklearn.neighbors.NearestNeighbors. The K distance is sorted in ascending order, a graph is drawn, and the distance corresponding to the inflection point is found as the eps parameter. The standardized feature vector and parameters are input, DBSCAN clustering is performed, and the label is output.

[0113] In one embodiment of the present invention, the fault diagnosis and classification module includes:

[0114] Calculate the characteristic center of each cluster and analyze its characteristic value distribution;

[0115] Build a library of known fault types, including fault type names, feature vectors, and similarity thresholds;

[0116] The similarity calculation is specifically as follows:

[0117]

[0118] Among them, S is the similarity, x j is the eigenvalue of the cluster center, y j is the characteristic value of a known fault type;

[0119] Define matching rules: If the similarity is less than the similarity threshold, the cluster is identified as a known fault type; otherwise, the cluster is identified as an unmatched cluster and analyzed.

[0120] Specifically, for each cluster, exclude the noise point -1, calculate the mean of its eigenvalues ​​as the feature center, analyze the feature center of each cluster, and observe its distribution in the three dimensions of signal reception, positioning, and hardware circuit; manually or based on historical data, define the name, feature vector, and similarity threshold of the known fault type, and save the known fault type library as a JSON file or database for subsequent use; use Euclidean distance to calculate the similarity between the cluster center and the feature vector of the known fault type, traverse each cluster center, calculate its similarity with the feature vectors of all known fault types, and output the matching result; Euclidean distance measures the degree of proximity between two feature vectors. The lower the similarity, the closer they are. Set the matching rule. If the similarity is less than the threshold, the cluster is identified as a known fault type, otherwise it is identified as an unknown fault type. The similarity threshold is set based on the distribution of the clustering results: calculate the similarity between all cluster centers and the feature vector of each known fault type, draw a similarity distribution graph, observe the similarity distribution of the known fault types, and set the threshold based on the distribution. For example, the maximum similarity of the known fault type is selected as the threshold to ensure that clusters of all known fault types can be correctly matched.

[0121] In one embodiment of the present invention, identifying the cluster as an unmatched cluster and analyzing the cluster includes:

[0122] Define the basic threshold as the global feature mean, calculate the current cluster density and global data density, and jointly construct the dynamic threshold. Specifically:

[0123]

[0124] Among them, T D is the dynamic threshold, T B is the basic threshold, D C is the current cluster density, D G is the global data density;

[0125] The KL divergence, time series cross similarity and Euclidean distance of the comprehensive feature distribution are used to assign weight coefficients and construct the fusion similarity, which is as follows:

[0126] S W =K d ×W1+S tc ×W2+D e ×W3

[0127] Among them, S W To fusion similarity, K d is the KL divergence, S tc is the temporal cross similarity, D e is the Euclidean distance, W1, W2 and W3 are the corresponding weight coefficients;

[0128] Calculate the sample feature statistics and draw a histogram to observe the feature distribution pattern; compare the cluster features with the features of the known fault model; calculate the Pearson correlation coefficient between the features; and make the first judgment: whether it is a new fault type;

[0129] Perform trend decomposition, sliding window detection, and time series pattern matching on time series data to determine whether it is a known fault; perform a second judgment: whether it is a time series fault or a new fault type;

[0130] Combine temperature, humidity, and pressure environmental data to perform correlation analysis, threshold detection, and joint clustering; and make a third judgment: whether it is an environment-related fault or a new fault type.

[0131] Specifically, the characteristic value distribution of the sample is analyzed, including calculating the statistical value of the cluster and drawing a histogram to observe the characteristic distribution form; the cluster characteristics are compared with the characteristics of the known fault model to find out the differences; the Pearson correlation coefficient between the characteristics is calculated to determine whether there is a multi-feature joint fault; the cluster characteristic analysis report is output, including statistical description, comparative analysis and correlation analysis, to make the first judgment: whether it is a new fault type; if the data contains a time dimension, the time series is decomposed into trend, seasonality and residual parts, abnormal changes are observed, and the moving average of the characteristics is calculated using a sliding window to detect mutation points. The cluster time series is matched with the time series model of the known fault type to determine whether it is a known fault; the time series analysis report is output, Including trend, seasonality, residual analysis and sliding window detection results, make a second judgment: whether it is a time series fault or a new fault type; combined with environmental data, including temperature, humidity and pressure, calculate the correlation coefficient between features and environmental data, judge whether there is an environmental, sensitivity fault, and check whether the environmental data exceeds the normal range, judge whether the fault is caused by extreme environmental conditions, jointly cluster the environmental data and features, and identify the environment-feature joint fault mode; output the environmental factor analysis report, including correlation analysis, threshold detection and joint clustering results, and make a third judgment: whether it is an environment-related fault or a new fault type, where W1, W2 and W3 are the corresponding weight coefficients, and the corresponding values ​​are 0.4, 0.4 and 0.3 respectively.

[0132] In one embodiment of the present invention, the new fault type identification and fault library update module includes:

[0133] Define identification rules: If the cluster cannot match a known fault type and meets the following conditions, it is represented as a new fault type;

[0134] The following conditions include that neither the dynamic threshold nor the fusion similarity calculation matches a known fault type, and the first, second, and third judgment results are all new fault types;

[0135] The new fault type information includes the fault type name, feature vector, similarity threshold and additional information, wherein the additional information includes time series pattern and environmental sensitivity description; the new fault type is added to the known fault type library;

[0136] At the same time, each time a new fault type is identified, the library of known fault types is automatically updated and the model is retrained.

[0137] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. Navigation hardware fault diagnosis system based on adaptive clustering algorithm, characterized by: Includes the following modules: Data acquisition module: collects operating status data from the satellite, obtains the quality data of the received signal and its own status data from the user receiver, and obtains monitoring data on the satellite's operating status from the ground control system; Data processing and feature extraction module: pre-processes the collected data, extracts key features from each type of processed data, and calculates the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value respectively; Adaptive algorithm clustering module: Constructs a three-dimensional feature vector from the signal comprehensive evaluation value, positioning comprehensive evaluation value, and hardware circuit comprehensive evaluation value, performs normalization, calculates the K distance of each sample, determines the EPS parameter through the ascending sorting graph, inputs the normalized feature vector and parameters, performs cluster analysis, and outputs the cluster label; Fault diagnosis and classification module: Based on the results of the adaptive clustering algorithm, similarity calculation and matching rules are defined, a library of known fault types is built to match the clustering results, and clusters that cannot be directly matched to known fault types are analyzed; New fault type identification and fault library update module: If the clustering cannot be matched, and the dynamic threshold, fusion similarity and multi-judgment results do not match, it is identified as a new fault type and updated to the known fault type library.

2. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1 is characterized in that: The data acquisition module includes: The operating status data includes various operating parameters of the satellite equipment: signal strength, frequency stability and attitude information; The quality data and status data include signal-to-noise ratio, bit error rate, and receiving sensitivity; the status information of the user receiver itself, including device temperature, battery level, and processor load; The monitoring data includes satellite orbit parameters, ground station communication link status and detection records of various satellite functions.

3. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The key features extracted from each type of processed data include: Establish a data processing server in the ground data processing center to receive data from all sources, clean the data, remove noise using a data filtering algorithm, and normalize the denoised data; Extracting key features from the processed data, including signal reception fault features, positioning error fault features, and hardware circuit fault features; Among them, the signal reception fault characteristics include signal strength related characteristics, signal-to-noise ratio characteristics and bit error characteristics; the positioning error fault characteristics include position deviation characteristics, position stability characteristics and satellite geometric distribution related characteristics; the hardware circuit fault characteristics include voltage and current characteristics, temperature characteristics and circuit characteristics related to signal quality.

4. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The comprehensive signal evaluation value includes: Count the relevant characteristics of the received signal strength within a preset time period, including the mean, variance and change rate, and take the average value to obtain the first signal evaluation value; Obtain the signal-to-noise ratio data of the signal, calculate the mean and standard deviation of the signal-to-noise ratio to obtain the signal-to-noise ratio variation coefficient as the second evaluation value of the signal; At the same time, the error characteristics of the received data are statistically analyzed, including the error rate, continuous error length, error distribution unevenness, and error interval. The error characteristics are weighted and added together to obtain the third evaluation value of the signal; The error distribution unevenness is obtained by collecting the error information received by the navigation system, classifying the error information according to actual needs, and calculating the probability of each error information by counting the number of occurrences. Specifically, Among them, R B is the unevenness of error distribution, p i is the probability of the i-th type of error, and n is the number of types of error; According to the importance of signal strength related features, signal-to-noise ratio features and error features to signal reception fault features, weight coefficients are assigned and weighted addition is performed to obtain the comprehensive signal evaluation value.

5. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The comprehensive positioning evaluation value includes: Calculate the average deviation between the user receiver position and the true position, as well as the degree of fluctuation of the deviation; calculate the rate of change of position over time for longitude, latitude, and altitude respectively, and take the average; detect whether the position jumps, and count the number of jumps to obtain the position jump frequency; count the number of currently visible satellites and calculate the geometric precision factor; Each feature is standardized; the standardized features are weighted and combined to obtain a comprehensive positioning evaluation value.

6. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The comprehensive evaluation value of the hardware circuit includes: The voltage and current characteristics include voltage mean, current mean, voltage standard deviation, and current standard deviation; the temperature characteristics include temperature mean and temperature standard deviation; the signal quality-related characteristics include radio frequency signal characteristics and intermediate frequency signal characteristics, wherein the radio frequency signal characteristics include radio frequency signal strength and signal-to-noise ratio, and the intermediate frequency signal characteristics include intermediate frequency signal amplitude and distortion; The features are normalized, and the normalized feature values ​​are weighted and added to obtain a comprehensive evaluation value of the hardware circuit.

7. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The adaptive algorithm clustering module includes: The comprehensive evaluation value of signal reception, comprehensive evaluation value of positioning and comprehensive evaluation value of hardware circuit are used as feature vectors, and each sample corresponds to a three-dimensional feature vector; The three comprehensive evaluation values ​​are standardized; Calculate the distance from each sample point to its nearest neighbor, that is, K distance, where K distance is equal to min_samples; draw an ascending sort graph of K distances and find the inflection point. The distance corresponding to the inflection point is eps; Input the standardized feature vector, parameters min_samples and eps, use Python's sklearn.cluster.DBSCAN for clustering, and output the cluster label; The cluster labels: -1 represents a noise point that cannot be classified into any cluster; 0 to n represent different cluster labels, and each cluster represents a potential fault type.

8. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The fault diagnosis and classification module includes: Calculate the characteristic center of each cluster and analyze its characteristic value distribution; Build a library of known fault types, including fault type names, feature vectors, and similarity thresholds; The similarity calculation is specifically as follows: Among them, S is the similarity, x j is the eigenvalue of the cluster center, y j is the characteristic value of a known fault type; Define matching rules: If the similarity is less than the similarity threshold, the cluster is identified as a known fault type; otherwise, the cluster is identified as an unmatched cluster and analyzed.

9. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 8, characterized in that: The step of identifying the cluster as an unmatched cluster and analyzing the cluster includes: Define the basic threshold as the global feature mean, calculate the current cluster density and global data density, and jointly construct the dynamic threshold. Specifically: Among them, T D is the dynamic threshold, T B is the basic threshold, D C is the current cluster density, D G is the global data density; The KL divergence, time series cross similarity and Euclidean distance of the comprehensive feature distribution are used to assign weight coefficients and construct the fusion similarity, which is as follows: S W =K d ×W1+S tc ×W2+D e ×W3 Among them, S W To fusion similarity, K d is the KL divergence, S tc is the temporal cross similarity, D e is the Euclidean distance, W1, W2 and W3 are the corresponding weight coefficients; Calculate the sample feature statistics and draw a histogram to observe the feature distribution pattern; compare the cluster features with the features of the known fault model; calculate the Pearson correlation coefficient between the features; and make the first judgment: whether it is a new fault type; Perform trend decomposition, sliding window detection, and time series pattern matching on time series data to determine whether it is a known fault; perform a second judgment: whether it is a time series fault or a new fault type; Combine temperature, humidity, and pressure environmental data to perform correlation analysis, threshold detection, and joint clustering; and make a third judgment: whether it is an environment-related fault or a new fault type.

10. The navigation hardware fault diagnosis system based on the adaptive clustering algorithm according to claim 1, characterized in that: The new fault type identification and fault library update module includes: Define identification rules: If the cluster cannot match a known fault type and meets the following conditions, it is represented as a new fault type; The following conditions include that neither the dynamic threshold nor the fusion similarity calculation matches a known fault type, and the first, second, and third judgment results are all new fault types; The new fault type information includes the fault type name, feature vector, similarity threshold and additional information, wherein the additional information includes time series pattern and environmental sensitivity description; the new fault type is added to the known fault type library; At the same time, each time a new fault type is identified, the library of known fault types is automatically updated and the model is retrained.

Images