Abnormal state auxiliary detection method for carrying out flight parameter data mining based on iForest

Abstract

The invention relates to a test flight monitoring and data processing technology, in particular to an abnormal state auxiliary detection method for carrying out flight parameter data mining based on iForest. The method comprises the following steps: collecting historical flight data, and carrying out data preprocessing; identifying and extracting the historical flight data by using an iForest algorithm to obtain first outlier information; calculating to obtain a first global correlation coefficient of the system in combination with a Spearman correlation theory, and obtaining a first local correlation coefficient of the system; calculating to obtain a system correlation degree threshold interval, and establishing a system correlation degree model; second outlier information in the data is extracted, a second global correlation coefficient of the system is calculated in combination with the Spearman correlation theory, and a second local correlation coefficient of the system is calculated; comparing the global correlation coefficient with a correlation threshold interval, and outputting auxiliary detection information; according to the method, the outliers in the data sequence are accurately identified, the abnormal fluctuation is effectively detected, and the sensitivity and the accuracy of abnormal detection are improved.

Description

Technical Field

[0001] The present invention relates to flight test monitoring and data processing technology, and in particular to an abnormal state auxiliary detection method based on iForest for flight parameter data mining. Background Art

[0002] With the rapid development of aviation technology, the intelligence level of aircraft has been continuously improved, resulting in the widespread distribution and increasing complexity of sensors equipped on aircraft, and a sharp increase in the types and amount of flight data generated. Flight data is not only an important basis for flight accident investigations, but also key information for evaluating aircraft performance, optimizing aircraft design, and improving flight safety. Therefore, how to efficiently and accurately process and analyze flight data and provide scientific guidance and intervention for aircraft development, flight testing, status monitoring, and maintenance has become an urgent problem that needs to be solved in the aviation field.

[0003] At present, most methods for processing flight parameter data rely on traditional data analysis technologies; however, these methods have many limitations when faced with massive amounts of data, such as large workload, low interpretation efficiency, and limited coverage of criteria. These problems restrict the effective use of flight data; in addition, existing methods can often only identify known fault modes, and have limited detection capabilities for new faults or unknown abnormal conditions, making it difficult to meet the high requirements of modern aviation for flight safety; on the other hand, existing flight parameter data processing methods also have the problem of a high false alarm rate; due to the complexity and variability of the flight environment, some non-fault factors may cause abnormal fluctuations in flight parameter data, thereby triggering false alarms; this not only increases the workload of maintenance personnel, but may also affect the timely discovery and handling of real faults.

[0004] There is a Chinese patent with publication number CN116521406A and publication date August 1, 2023. The patent name is "A method for detecting anomalies in flight parameter data of aircraft engines that are not exceeding the limit based on the residual gate GRU-VAE model". The specific technical solution is as follows: The invention discloses a method for detecting anomalies in flight parameter data of aircraft engines that are not exceeding the limit based on the residual gate GRU-VAE model. In the model training stage, a historical flight parameter data set representing the engine performance baseline is used to train the residual gate GRU-VAE flight parameter anomaly detection model, and all flight parameter sequences in the data set are reconstructed. According to the probability distribution obeyed by the state point vector reconstruction loss, the corresponding reconstruction loss threshold is determined under a given confidence level; in the anomaly detection stage, the model detects real-time flight parameter data, and regards the state points whose reconstruction loss exceeds the threshold as abnormal state points, and further determines the flight parameter sequences whose abnormal state point ratio exceeds the ratio threshold as abnormal flight parameter sequences.

[0005] The above-mentioned patent uses historical flight parameter data to train the residual gate GRU-VAE model, and detects abnormal flight parameter data through the model. However, the training of the residual gate GRU-VAE model is too complicated, and the correlation and regularity of the flight parameter data are not studied, so the detection accuracy of the abnormal flight parameter data is not accurate enough. Summary of the Invention

[0006] In order to solve the problems existing in the prior art, the present invention provides an abnormal state auxiliary detection method based on iForest flight parameter data mining, which constructs a system correlation model based on data mining, and performs system state analysis through the correlation model to achieve efficient processing of flight parameter data and accurate identification of abnormal fluctuations.

[0007] In order to achieve the above technical effects, the technical solutions of this application are as follows:

[0008] like Figure 1 As shown, an abnormal state auxiliary detection method based on iForest flight parameter data mining includes the following specific method steps:

[0009] Step 1: Collect historical flight data and perform data preprocessing on the collected historical flight data;

[0010] Step 2: Use the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing;

[0011] Step 3: Using the pre-processed historical flight data and the Spearman correlation theory, the first global correlation coefficient of the system is calculated. The first outlier information is used to calculate the first local correlation coefficient of the system. The variance of the first global correlation coefficient and the first local correlation coefficient is then minimized.

[0012] Step 4: Calculate a system correlation threshold interval based on the obtained first global correlation coefficient and the first local correlation coefficient, thereby establishing a system correlation model; the system correlation threshold interval includes a global correlation threshold interval and a local correlation threshold interval;

[0013] Step 5: Collect flight data to be tested and perform data preprocessing on the flight data to be tested; use the iForest algorithm to identify and extract the second outlier information from the preprocessed flight data to be tested, and then use the Spearman correlation theory to calculate the second global correlation coefficient of the system. Use the second outlier information to calculate the second local correlation coefficient of the system;

[0014] Step 6: Compare the second global correlation coefficient with the global correlation threshold interval, and compare the second local correlation coefficient with the local correlation threshold interval, and output auxiliary detection information based on the final comparison result.

[0015] Furthermore, the specific method of data preprocessing is: set 60 sample data as a window and use the median to perform filtering; the data set before preprocessing can be expressed as X = {x1, x2, ..., x m}, where m is the amount of data collected, and the preprocessing is specifically expressed as:

[0016]

[0017] Where, k is the window data size, and the present invention sets k to 60; X t is the median of the window; t is the number of windows; median{} is the median filtering process.

[0018] Preprocessed dataset X per It can be expressed as:

[0019] X per ={X1, X2, ..., X t}.

[0020] Furthermore, the specific steps of using the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing are as follows:

[0021] Step a: From the preprocessed dataset X per Randomly select ψ points from as samples and place them at the root node of an isolated tree;

[0022] Step b: Randomly select a dimension and randomly generate a cutting point P within the data range of the current node;

[0023] Step c: Use the cutting point P to form a hyperplane and divide the current node data space into two subspaces; put the points with a dimension less than P into the left branch of the current node, and the points with a dimension greater than P into the right branch;

[0024] Step d: Repeat steps b to c to continue building new leaf nodes until data segmentation is no longer possible or the number of segmentations reaches log2 ψ When , the segmentation process is stopped;

[0025] Step e: After stopping the segmentation process, start calculating the normalization factor;

[0026] Step f: Calculate the path length of each sample data according to the obtained normalization factor, and obtain the average path length of each sample data;

[0027] Step g: Calculate the abnormal score of each sample data according to the average path length of each sample data, and obtain the first outlier information from the data sample set determined to be an outlier.

[0028] Furthermore, the specific formula for calculating the normalization factor in step e is as follows:

[0029]

[0030] Where c(ψ) is the normalization factor and ξ is the Euler constant.

[0031] Furthermore, the path length calculation formula for each sample data in step f is as follows:

[0032] h(x)=c(ψ)+μ;

[0033] Where h(x) is the path length of the sample data x from the root node to the leaf node in a single isolated tree; μ is the number of edges in the path from the root node to the leaf node;

[0034] The formula for calculating the average path length of each sample data is as follows:

[0035]

[0036] Where E(x) is the average path length of sample data x; r is the total number of isolated trees.

[0037] Furthermore, the specific solution formula for step g is as follows:

[0038]

[0039] Where s(x) is the anomaly score of the sample data x; when E(x)→0, s→1, the data sample is considered an outlier; when E(x)→ψ-1, s→0, the data is considered normal; when E(hx)→c(ψ), s→0.5, the data is considered to have no obvious anomalies.

[0040] Furthermore, in step 3, the specific formula for calculating the first global correlation coefficient of the system using the historical flight data after data preprocessing combined with the Spearman correlation theory is:

[0041]

[0042] Where R i and S i The rank of observation i after sorting the two data sets respectively; and are the average ranks of the two data sets respectively; N is the total number of observations; d i =Ri -S i , represents the difference in rank between observations i in two data sets; ρ sa-i，j is the global correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

[0043] The specific formula for calculating the first local correlation coefficient of the system using the first outlier information is as follows:

[0044]

[0045] Where, ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

[0046] Furthermore, the coefficient set of the first global correlation coefficient and the coefficient set of the first local correlation coefficient of the n-th flight are specifically expressed as follows:

[0047] ρ sa-n =[ρ sa-n-1，2 , ρ sa-n-1，3 ,…,ρ sa-n-i，j ];

[0048] ρ sl-n =[ρ sl-n-1,2 , ρ sl-n-1,3 ,…,ρ sl-n-i，j ];

[0049] Where, ρ sa-n is the first global correlation coefficient of the nth flight; n is the number of historical flight data collected; ρ sa-i，j is the first global correlation coefficient between the i-th system and the n-th system in the n-th flight, and i≠j; ρ sl-n is the first local correlation coefficient of the nth flight; ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the,-th system in the n-th flight, and i≠j.

[0050] Furthermore, based on the obtained first global correlation coefficient and the first local correlation coefficient, the specific method for calculating the system correlation threshold interval is: defining the maximum value of the first global correlation coefficient as the upper limit value of the global correlation threshold interval, and defining the minimum value of the first global correlation coefficient as the lower limit value of the global correlation threshold interval; defining the maximum value of the first local correlation coefficient as the upper limit value of the local correlation threshold interval, and defining the minimum value of the first local correlation coefficient as the lower limit value of the local correlation threshold interval.

[0051] Furthermore, the specific expression of the global correlation threshold interval is:

[0052] ρa-i，j =[ρ aH-i,j ,ρ aL-i，j ]=[max(ρ sa-n-i，j ), min(ρ sa-n-i，j )];

[0053] Where, ρ a-i，j is the first global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the j-th system, and i≠j.

[0054] Furthermore, the specific expression of the local correlation threshold interval is:

[0055] ρ l-i，j =[ρ lH-i，j , ρ lL-i，j ]=[max(ρ sl-n-i，j ), min(ρ sl-n-i，j )];

[0056] Where, ρ l-i，j is the local correlation threshold interval between the i-th system and the j-th system, and i≠j; ρ lH-i，j and ρ lL-i，j are the upper and lower limits of the local correlation threshold interval between the i-th system and the j-th system, respectively, and i≠j.

[0057] Furthermore, the flight data to be tested after data preprocessing is used to identify and extract outlier information from the data using the iForest algorithm. The second global correlation coefficient of the system is calculated using the Spearman correlation theory, and the second local correlation coefficient of the system is calculated using the outlier information. The specific expression is as follows:

[0058] ρ sa =[ρ sa-1，2 , ρ sa-1，3 ,…,ρ sa-i,j ];

[0059] ρ sl =[ρ sl-1，2 , ρ sl-1，3 ,…,ρ sl-i,j ];

[0060] Where, ρ sa is the second global correlation coefficient of the system; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ sl is the second local correlation coefficient of the system; ρ sl-i，jis the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0061] Furthermore, the final comparison result in step six includes Class I priority faults, Class II priority faults and Class III priority faults, and the processing priority order decreases step by step in the order of Class I priority faults, Class II priority faults and Class III priority faults; the auxiliary detection information includes situation 1 and situation 2; the Class I priority fault corresponds to situation 1, and the auxiliary detection information of situation 1 is output; the Class II priority fault corresponds to situation 2, and the auxiliary detection information of situation 2 is output; the Class III priority fault corresponds to situation 3, and no auxiliary detection information is output.

[0062] Furthermore, the Class I priority fault is a fault that should be handled but is not handled. The specific expression for determining the occurrence of the Class I priority fault is as follows:

[0063] ρ aH-i，j ·ρ aL-i,j >0;

[0064] ρ sa-i，j <ρ aL-i，j ;

[0065] Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0066] The Class II priority fault is an anomaly. The specific expression for determining the occurrence of a Class II priority fault is as follows:

[0067] ρ aH-i，j ·ρ aL-i，j >0;

[0068] ρ sa-i，j >ρ aH-i，j ;

[0069] Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0070] The Class III priority fault is normal. When the second global correlation meets the global correlation threshold, the second local correlation meets the local correlation threshold, and it does not belong to Class I priority fault and Class II priority fault, it is determined to be a Class II priority fault.

[0071] Furthermore, the case 2 is specifically to arrange the outlier anomaly scores s(x) in descending order, select the top 20 highest-scoring samples, trace back to the time point when the anomaly sample occurred based on the collected original data, and output the auxiliary detection information of case 2.

[0072] Furthermore, the auxiliary detection information of situation 1 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient thresholds, actual correlation coefficients and judgment priority; the auxiliary detection information of situation 2 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient thresholds, actual correlation coefficients, abnormal time points and abnormal parameter waveforms.

[0073] According to the above technical solution, the beneficial effects of this application are as follows:

[0074] 1. The present invention uses the iForest algorithm to mine flight parameter data, accurately identifying outliers in the data sequence and effectively detecting abnormal fluctuations, thereby improving the sensitivity and accuracy of anomaly detection.

[0075] 2. This invention combines the Spearman correlation theory to construct a system correlation model, further revealing the inherent correlation and regularity between aircraft system parameters, and providing a more comprehensive and accurate basis for auxiliary detection of abnormal conditions.

[0076] 3. The present invention generates auxiliary detection information based on the constructed system correlation model and combines it with human intervention to accurately identify abnormal fluctuations in subsystem flight parameter data in a short period of time. This method not only improves the efficiency and accuracy of anomaly detection, but also provides a new technical means for aviation safety monitoring and early warning, which is of great significance in reducing the risk of flight accidents and improving the level of aviation safety.

[0077] 4. The method of the present invention has important theoretical value and practical significance in improving the efficiency and accuracy of flight data analysis, revealing the intrinsic correlation of parameters, assisting in anomaly detection, and improving the level of aviation safety. It provides a new solution for data processing and analysis in the aviation field. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0080] Example 1

[0081] An abnormal state auxiliary detection method based on iForest flight parameter data mining includes the following specific method steps:

[0082] Step 1: Collect historical flight data and perform data preprocessing on the collected historical flight data; data preprocessing is a mature existing technology in this field.

[0083] Step 2: Use the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing;

[0084] Step 3: Use the pre-processed historical flight data in combination with the Spearman correlation theory to calculate the first global correlation coefficient of the system, and use the first outlier information to calculate the first local correlation coefficient of the system. Then, minimize the variance of the first global correlation coefficient and the first local correlation coefficient.

[0085] Step 4: Calculate a system correlation threshold interval based on the obtained first global correlation coefficient and the first local correlation coefficient, thereby establishing a system correlation model based on data mining to reveal the correlation and regularity between flight parameters; the system correlation threshold interval includes a global correlation threshold interval and a local correlation threshold interval. Steps 1 to 4 are for constructing a system correlation model based on data mining;

[0086] Step 5: Collect flight data to be tested and perform data preprocessing on the flight data to be tested; use the iForest algorithm to identify and extract the second outlier information from the preprocessed flight data to be tested, and then use the Spearman correlation theory to calculate the second global correlation coefficient of the system. The second outlier information is then used to calculate the second local correlation coefficient of the system.

[0087] Step 6: Compare the second global correlation coefficient with the global correlation threshold interval, and compare the second local correlation coefficient with the local correlation threshold interval. Based on the final comparison result, auxiliary detection information is output. Through auxiliary detection information and manual intervention, abnormal fluctuations in subsystem flight parameter data can be accurately identified in a short time, thereby improving the efficiency and accuracy of flight data analysis and interpretation; Steps 5 to 6 are for using the correlation model to perform system status analysis.

[0088] Example 2

[0089] An abnormal state auxiliary detection method based on iForest flight parameter data mining includes the following specific method steps:

[0090] Step 1: Collect historical flight data (collect n times of historical flight data), and preprocess the collected historical flight data to retain data feature information and reduce the data volume; data preprocessing is a mature existing technology in this field.

[0091] Step 2: Use the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing;

[0092] Step 3: Use the pre-processed historical flight data combined with the Spearman correlation theory to calculate the first global correlation coefficient of the system, and use the first outlier information to calculate the first local correlation coefficient of the system. Then, minimize the variance of the first global correlation coefficient and the first local correlation coefficient to reduce the influence of singular values.

[0093] Step 4: Calculate a system correlation threshold interval based on the obtained first global correlation coefficient and the first local correlation coefficient, thereby establishing a system correlation model based on data mining to reveal the correlation and regularity between flight parameters; the system correlation threshold interval includes a global correlation threshold interval and a local correlation threshold interval. Steps 1 to 4 are for constructing a system correlation model based on data mining;

[0094] Step 5: Collect the flight data to be tested and perform data preprocessing on the flight data to be tested; use the iForest algorithm to identify and extract the second outlier information in the preprocessed flight data to be tested, and then use the Spearman correlation theory to calculate the second global correlation coefficient of the system, and use the second outlier information to calculate the second local correlation coefficient of the system. Through auxiliary detection information and manual intervention, abnormal fluctuations in the subsystem flight parameter data can be accurately identified in a short time, thereby improving the efficiency and accuracy of flight data analysis and interpretation.

[0095] Step 6: Compare the second global correlation coefficient with the global correlation threshold interval, and compare the second local correlation coefficient with the local correlation threshold interval, and output auxiliary detection information based on the final comparison result; Steps 5 to 6 are to use the correlation model to perform system status analysis.

[0096] The specific method of data preprocessing is: set 60 sample data as a window and use the median for filtering. Using the median for filtering retains the data feature information and reduces the data volume, thereby improving the calculation speed. The data set before preprocessing can be expressed as X = {x1, x2, ..., x m}, where m is the amount of data collected, and the preprocessing is specifically expressed as:

[0097]

[0098] Where, k is the window data size, and the present invention sets k to 60; X t is the median of the window; t is the number of windows; median{} is the median filtering process.

[0099] Preprocessed dataset X per It can be expressed as:

[0100] X per ={X1, X2, ..., X t}.

[0101] Example 3

[0102] Based on Example 2, the specific steps of using the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing are as follows:

[0103] Step a: From the preprocessed dataset X per Randomly select ψ points from as samples and place them at the root node of an isolated tree;

[0104] Step b: Randomly select a dimension and randomly generate a cutting point P within the data range of the current node;

[0105] Step c: Use the cutting point P to form a hyperplane and divide the current node data space into two subspaces; put the points with a dimension less than P into the left branch of the current node, and the points with a dimension greater than P into the right branch;

[0106] Step d: Repeat steps b to c to continue building new leaf nodes until data segmentation is no longer possible or the number of segmentations reaches log2 ψ When , the segmentation process is stopped;

[0107] Step e: After stopping the segmentation process, start calculating the normalization factor;

[0108] Step f: Calculate the path length of each sample data according to the obtained normalization factor, and obtain the average path length of each sample data;

[0109] Step g: Calculate the abnormal score of each sample data according to the average path length of each sample data, and obtain the first outlier information from the data sample set determined to be an outlier.

[0110] The specific formula for calculating the normalization factor in step e is as follows:

[0111]

[0112] Where c(ψ) is the normalization factor and ξ is the Euler constant.

[0113] Step f calculates the path length of each sample data as follows:

[0114] h(x)=c(ψ)+μ;

[0115] Where h(x) is the path length of the sample data x from the root node to the leaf node in a single isolated tree; μ is the number of edges in the path from the root node to the leaf node;

[0116] The formula for calculating the average path length of each sample data is as follows:

[0117]

[0118] Where E(x) is the average path length of sample data x; r is the total number of isolated trees.

[0119] The specific solution formula for step g is as follows:

[0120]

[0121] Where s(x) is the anomaly score of the sample data x; when E(x)→0, s→1, the data sample is considered to be an outlier and also an abnormal point of the data; when E(x)→ψ-1, s→0, the data is considered to be normal; when E(hx)→c(ψ), s→0.5, the data is considered to have no obvious abnormalities.

[0122] Example 4

[0123] Based on Example 3, in step 3, the specific formula for calculating the first global correlation coefficient of the system using the historical flight data after data preprocessing combined with the Spearman correlation theory is:

[0124]

[0125] Where R i and S i The rank of observation i after sorting the two data sets respectively; and are the average ranks of the two data sets respectively; N is the total number of observations; di =R i -S i , represents the difference in rank between observations i in two data sets; ρ sa-i，j is the global correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

[0126] The specific formula for calculating the first local correlation coefficient of the system using the first outlier information is as follows:

[0127]

[0128] Where, ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

[0129] The coefficient set of the first global correlation coefficient and the coefficient set of the first local correlation coefficient of the n-th flight are specifically expressed as follows:

[0130] ρ sa-n =[ρ sa-n-1，2 , ρ sa-n-1，3 ,…,ρ sa-n-i，j ];

[0131] ρ sl-n =[ρ sl-n-1,2 , ρ sl-n-1,3 ,…,ρ sl-n-i，j ];

[0132] Where, ρ sa-n is the first global correlation coefficient of the nth flight; n is the number of historical flight data collected; ρ sa-i，j is the first global correlation coefficient between the i-th system and the n-th system in the n-th flight, and i≠j; ρ sl-n is the first local correlation coefficient of the nth flight; ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the,-th system in the n-th flight, and i≠j.

[0133] Based on the obtained first global correlation coefficient and the first local correlation coefficient, the specific method of calculating the system correlation threshold interval is: defining the maximum value of the first global correlation coefficient as the upper limit value of the global correlation threshold interval, and defining the minimum value of the first global correlation coefficient as the lower limit value of the global correlation threshold interval; defining the maximum value of the first local correlation coefficient as the upper limit value of the local correlation threshold interval, and defining the minimum value of the first local correlation coefficient as the lower limit value of the local correlation threshold interval.

[0134] The specific expression of the global correlation threshold interval is:

[0135] ρa-i，j =[ρ aH-i,j ,ρ aL-i，j ]=[max(ρ sa-n-i，j ), min(ρ sa-n-i，j )];

[0136] Where, ρ a-i，j is the first global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the j-th system, and i≠j.

[0137] The specific expression of the local correlation threshold interval is:

[0138] ρ l-i，j =[ρ lH-i，j , ρ lL-i，j ]=[max(ρ sl-n-i，j ), min(ρ sl-n-i，j )];

[0139] Where, ρ l-i，j is the local correlation threshold interval between the i-th system and the j-th system, and i≠j; ρ lH-i，j and ρ lL-i，j are the upper and lower limits of the local correlation threshold interval between the i-th system and the j-th system, respectively, and i≠j.

[0140] The flight data to be tested after data preprocessing is used to identify and extract outlier information in the data through the iForest algorithm. Then, the second global correlation coefficient of the system is calculated by combining it with the Spearman correlation theory. The specific expression of the second local correlation coefficient of the system is calculated using the outlier information as follows:

[0141] ρ sa =[ρ sa-1，2 , ρ sa-1，3 ,…,ρ sa-i,j ];

[0142] ρ sl =[ρ sl-1，2 , ρ sl-1，3 ,…,ρ sl-i,j ];

[0143] Where, ρ sa is the second global correlation coefficient of the system; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ sl is the second local correlation coefficient of the system; ρ sl-i，jis the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0144] Example 5

[0145] Based on Example 4, the final comparison result in step 6 includes Class I priority faults, Class II priority faults and Class III priority faults, and the processing priority order is gradually reduced in the order of Class I priority faults, Class II priority faults and Class III priority faults; the auxiliary detection information includes Case 1 and Case 2; the Class I priority fault corresponds to Case 1, and the auxiliary detection information of Case 1 is output; the Class II priority fault corresponds to Case 2, and the auxiliary detection information of Case 2 is output; the Class III priority fault corresponds to Case 3, and no auxiliary detection information is output; Case 2 specifically arranges the abnormal scores s(x) of the outliers in descending order, selects the top 20 highest-scoring samples, and traces back to the occurrence of the abnormal samples based on the collected original data. At the time point, the auxiliary detection information of Case 2 is output; the auxiliary detection information outputted by Case 1 and Case 2 is shown in the auxiliary detection information in Table 1 below, the auxiliary detection information of Case 1 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient threshold, actual correlation coefficient and interpretation priority; the auxiliary detection information of Case 2 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient threshold, actual correlation coefficient, abnormal time point and abnormal parameter waveform; Case 1 means that the second local correlation coefficient meets the local correlation threshold interval, but the second global correlation coefficient does not meet the global correlation threshold interval; Case 2 means that the second global correlation coefficient meets the global correlation threshold interval, but the second local correlation coefficient does not meet the local correlation threshold interval.

[0146] Table 1 Auxiliary detection information

[0147]

[0148] Class I priority faults are those that should be handled but are not. The specific expression for determining the occurrence of Class I priority faults is as follows:

[0149] ρ aH-i，j ·ρ aL-i,j >0;

[0150] ρ sa-i，j <ρ aL-i，j ;

[0151] Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0152] Class II priority faults are abnormal. The specific expression for determining the occurrence of Class II priority faults is as follows:

[0153] ρ aH-i，j ·ρ aL-i，j >0;

[0154] ρ sa-i，j >ρ aH-i，j ;

[0155] Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

[0156] Class III priority faults are normal. When the second global correlation meets the global correlation threshold, the second local correlation meets the local correlation threshold, and it does not belong to Class I priority faults and Class II priority faults, it is determined to be a Class III priority fault.

[0157] The above description is a detailed description of the preferred feasible embodiment of the present application, but the embodiment is not intended to limit the scope of the patent application of the present application. Any equivalent changes or modifications completed under the technical spirit suggested by the present application should fall within the scope of the patent covered by the present application.

Claims

1. An abnormal state auxiliary detection method based on iForest flight parameter data mining includes the following specific method steps: Step 1: Collect historical flight data and perform data preprocessing on the collected historical flight data; Step 2: Use the iForest algorithm to identify and extract the first outlier information from the historical flight data after data preprocessing; Step 3: Using the pre-processed historical flight data and the Spearman correlation theory, the first global correlation coefficient of the system is calculated. The first outlier information is used to calculate the first local correlation coefficient of the system. The variance of the first global correlation coefficient and the first local correlation coefficient is then minimized. Step 4: Calculate a system correlation threshold interval based on the obtained first global correlation coefficient and the first local correlation coefficient, thereby establishing a system correlation model; the system correlation threshold interval includes a global correlation threshold interval and a local correlation threshold interval; Step 5: Collect flight data to be tested and perform data preprocessing on the flight data to be tested; use the iForest algorithm to identify and extract the second outlier information from the preprocessed flight data to be tested, and then use the Spearman correlation theory to calculate the second global correlation coefficient of the system. Use the second outlier information to calculate the second local correlation coefficient of the system; Step 6: Compare the second global correlation coefficient with the global correlation threshold interval, and compare the second local correlation coefficient with the local correlation threshold interval, and output auxiliary detection information based on the final comparison result.

2. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 1 is characterized in that: The specific method of data preprocessing is: set 60 sample data as a window and use median to perform filtering; the data set before preprocessing can be expressed as X = {x1, x2, ..., x m }, where m is the amount of data collected, and the preprocessing is specifically expressed as: Where, k is the window data size, and the present invention sets k to 60; X t is the median of the window; t is the number of windows; median{} is the median filtering process; Preprocessed dataset X per It can be expressed as: X per ={X1,X2,...,X t }。 3. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 2 is characterized in that: The specific steps for using the iForest algorithm to identify and extract the first outlier information from the pre-processed historical flight data are as follows: Step a: From the preprocessed dataset X per Randomly select ψ points from as samples and place them at the root node of an isolated tree; Step b: Randomly select a dimension and randomly generate a cutting point P within the data range of the current node; Step c: Use the cutting point P to form a hyperplane to divide the current node data space into two subspaces; Put the points with dimension less than P into the left branch of the current node, and the points with dimension greater than P into the right branch; Step d: Repeat steps b to c to continue building new leaf nodes until data segmentation is no longer possible or the number of segmentations reaches log2 ψ When , the segmentation process is stopped; Step e: After stopping the segmentation process, start calculating the normalization factor; Step f: Calculate the path length of each sample data according to the obtained normalization factor, and obtain the average path length of each sample data; Step g: Calculate the abnormal score of each sample data according to the average path length of each sample data, and obtain the first outlier information from the data sample set determined to be an outlier.

4. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 3 is characterized by: The specific formula for calculating the normalization factor in step e is as follows: Where c(ψ) is the normalization factor and ξ is the Euler constant.

5. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 3 is characterized in that: Step f calculates the path length of each sample data as follows: h(x)=c(ψ)+μ; Where h(x) is the path length of the sample data x from the root node to the leaf node in a single isolated tree; μ is the number of edges in the path from the root node to the leaf node; The formula for calculating the average path length of each sample data is as follows: Where E(x) is the average path length of sample data x; r is the total number of isolated trees.

6. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 3 is characterized by: The specific solution formula for step g is as follows: Where s(x) is the anomaly score of the sample data x; when E(x)→0, s→1, the data sample is considered an outlier; when E(x)→ψ-1, s→0, the data is considered normal; when E(hx)→c(ψ), s→0.5, the data is considered to have no obvious anomalies.

7. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 1 is characterized in that: In step 3, the specific formula for calculating the first global correlation coefficient of the system using the historical flight data after data preprocessing combined with the Spearman correlation theory is: Where R i and S i The rank of observation i after sorting the two data sets respectively; and are the average ranks of the two data sets respectively; N is the total number of observations; d i =R i -S i , represents the difference in rank between observations i in two data sets; ρ sa-i，j is the global correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j; The specific formula for calculating the first local correlation coefficient of the system using the first outlier information is as follows: Where, ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

8. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 7 is characterized in that: The coefficient set of the first global correlation coefficient and the coefficient set of the first local correlation coefficient of the n-th flight are specifically expressed as follows: r sa-n =[ρ sa-n-1，2 ,r sa-n-1，3 ,…,r sa-n-i，j ]; r sl-n =[ρ sl-n-1,2 ,r sl-n-1,3 ,…,r sl-n-i,j ]; Where, ρ sa-n is the first global correlation coefficient of the nth flight; n is the number of historical flight data collected; ρ sn-i，j is the first global correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j; ρ sl-n is the first local correlation coefficient of the nth flight; ρ sl-n-i，j is the first local correlation coefficient between the i-th system and the j-th system in the n-th flight, and i≠j.

9. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 1 is characterized in that: Based on the obtained first global correlation coefficient and the first local correlation coefficient, the specific method of calculating the system correlation threshold interval is: defining the maximum value of the first global correlation coefficient as the upper limit value of the global correlation threshold interval, and defining the minimum value of the first global correlation coefficient as the lower limit value of the global correlation threshold interval; defining the maximum value of the first local correlation coefficient as the upper limit value of the local correlation threshold interval, and defining the minimum value of the first local correlation coefficient as the lower limit value of the local correlation threshold interval.

10. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 9 is characterized in that: The specific expression of the global correlation threshold interval is: r a-i，j =[ρ aH-i,j ,r aL-i，j ]=[max(ρ sa-n-i，j ),min(ρ sa-n-i，j )]; Where, ρ a-i，j is the first global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the j-th system, and i≠j.

11. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 9 is characterized in that: The specific expression of the local correlation threshold interval is: r l-i，j =[ρ lH-i，j ,r lL-i，j ]=[max(ρ sl-n-i，j ),min(ρ sl-n-i，j )]; Where, ρ l-i，j is the local correlation threshold interval between the i-th system and the j-th system, and i≠j; ρ lH-i，j and ρ lL-i，j are the upper and lower limits of the local correlation threshold interval between the i-th system and the j-th system, respectively, and i≠j.

12. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 1 is characterized in that: The flight data to be tested after data preprocessing is used to identify and extract outlier information in the data through the iForest algorithm. Then, the second global correlation coefficient of the system is calculated by combining it with the Spearman correlation theory. The specific expression of the second local correlation coefficient of the system is calculated using the outlier information as follows: r sa =[ρ sa-1，2 ,r sa-1，3 ,…,r sa-i,j ]; r sl =[ρ sl-1，2 ,r sl-1，3 ,…,r sl-i,j ]; Where, ρ sa is the second global correlation coefficient of the system; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j; ρ sl is the second local correlation coefficient of the system; ρ sl-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j.

13. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 1 is characterized in that: The final comparison result in step six includes Class I priority faults, Class II priority faults and Class III priority faults, and the processing priority order decreases step by step in the order of Class I priority faults, Class II priority faults and Class III priority faults; the auxiliary detection information includes situation 1 and situation 2; the Class I priority fault corresponds to situation 1, and the auxiliary detection information of situation 1 is output; the Class II priority fault corresponds to situation 2, and the auxiliary detection information of situation 2 is output; the Class III priority fault corresponds to situation 3, and no auxiliary detection information is output.

14. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 13 is characterized in that: The Class I priority fault is a fault that should be handled but is not handled. The specific expression for determining the occurrence of a Class I priority fault is as follows: r aH-i，j ·r aL-i,j >0; r sa-i，j <p aL-i，j ; Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j; The Class II priority fault is an anomaly. The specific expression for determining the occurrence of a Class II priority fault is as follows: r aH-i，j ·r aL-i,j >0; r sa-i，j >r aH-i，j ; Where, ρ aH-i，j represents the upper limit of the global correlation threshold interval of the i-th system, and i≠j; ρ aL-i，j represents the lower limit of the global correlation threshold interval of the jth system, and i≠j; ρ sa-i，j is the second global correlation coefficient between the i-th system and the j-th system, and i≠j; The Class III priority fault is normal. When the second global correlation meets the global correlation threshold, the second local correlation meets the local correlation threshold, and it does not belong to Class I priority fault and Class II priority fault, it is determined to be a Class III priority fault.

15. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 13, characterized in that: The specific case 2 is to arrange the outlier anomaly scores s(x) in descending order, select the top 20 highest-scoring samples, trace back to the time point when the anomaly sample occurred based on the collected original data, and output the auxiliary detection information of case 2.

16. The abnormal state auxiliary detection method based on iForest flight parameter data mining according to claim 13, characterized in that: The auxiliary detection information for situation 1 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient threshold, actual correlation coefficient and interpretation priority; the auxiliary detection information for situation 2 includes flight sorties, associated subsystems, associated indicators, reference correlation coefficient threshold, actual correlation coefficient, abnormal time point and abnormal parameter waveform.

Images