An unmanned aerial vehicle variable-density flight data working condition recognition method based on mutual neighbors

Abstract

The application discloses a method for recognizing the working condition of unmanned aerial vehicle variable-density flight data based on mutual near neighbors, inputs a new data point and a data set, obtains k near neighbor data and working conditions, constructs an extended data set if the working condition is equal to 1, determines the new data point as a local outlier or outputs the corresponding working condition according to the value of the mutual near neighbor indicator function, randomly selects a working condition if the working condition is not equal to 1, constructs an extended data set, determines the new data point as a local outlier if the number of mutual near neighbors of the new data point is 0, and assigns a value to the position sequence indicator function if the number of mutual near neighbors of the new data point is not 0, and outputs the working condition of the new data point. The application can be applied to unmanned aerial vehicle variable-density flight data. The working condition discrimination method based on mutual near neighbors can quickly determine the working condition of the new data in the variable-density data, and extends the working condition recognition method from uniform distribution data to variable-density data.

Description

TECHNICAL FIELD

[0001] The method belongs to the field of pattern recognition, and particularly relates to a variable-density data working condition recognition method based on mutual neighbors. BACKGROUND

[0002] In industrial production processes, medical examination and detection, financial transactions, biological information acquisition and network information processing, the acquired data often presents different categories, the data in the same category presents high similarity under a certain rule, and the data between different categories presents high dissimilarity, that is, presents multi-working condition characteristics. With the evolution of the process, new data needs to be combined with historical data to recognize the working condition, and the working condition recognition accuracy directly determines the effects of industrial equipment health management, medical examination, financial security protection, biological information attribution recognition and network anti-invasion.

[0003] Common working condition recognition methods include model-based methods, such as establishing a working condition recognition model based on a support vector machine or a neural network; deep learning-based methods, such as using a deep neural network to extract object features and establishing a working condition mapping relationship; similarity-based methods, such as using different distance metrics or other similarity metrics to recognize the working condition of data; and probability-based methods and information entropy-based methods.

[0004] However, in actual situations, the historical data distribution is often non-uniform, that is, presents a variable-density characteristic. Common working condition recognition methods are difficult to adapt to the variable-density characteristics of data, resulting in low algorithm accuracy. In addition, when new data is an outlier, especially a local outlier, common working condition recognition methods will incorrectly recognize it as one of the historical working conditions, and therefore do not have outlier detection capability. SUMMARY

[0005] The present application aims to solve the above problems in the prior art, and provides a variable-density flight data working condition recognition method for a UAV based on mutual neighbors, which can recognize local outliers and global outliers.

[0006] The above technical problems of the present application are mainly solved by the following technical scheme:

[0007] A variable-density flight data working condition recognition method for a UAV based on mutual neighbors, comprising the following steps,

[0008] Step 1, inputting new data points d of UAV flight data p and a UAV flight data set D, the UAV flight data set D being a known data set and containing l working condition sub-data sets D1-D l , C1-C l are working condition sub-data sets D1-D lCorresponding working condition class label,

[0009] Let C() denote the working condition class label of a UAV flight data point, and let k D () denote the k nearest neighbor data set of a UAV flight data point in the UAV flight data set D, let lk D () denote the working condition contained in the k nearest neighbor data set of a UAV flight data point in the UAV flight data set D, and let |lk D ()| denote the number of working conditions contained in the k nearest neighbor data set of a UAV flight data point in the UAV flight data set D,

[0010] Step 2, the UAV flight data set D and the newly added data point d p are input into the k nearest neighbor algorithm to obtain the k nearest neighbor data set k p (d D ) of the newly added data point d p in the UAV flight data set D, p and the working condition lk D (d p ) contained in the k nearest neighbor data set of the newly added data point d p in the UAV flight data set D,

[0011] Step 3, as shown in the figure, if the number of working conditions to which the k nearest neighbor data set of the newly added data point d D in the UAV flight data set D belongs is |lk p (d p )| = 1, that is, all data points in the k nearest neighbor data set k D (d p ) of the newly added data point d m belong to the same working condition C p , then the newly added data point d m is added to the corresponding working condition sub-data set D D , m ∈ lk p (d m ), to form an extended UAV flight data set D i ', otherwise, go to step 6,

[0012] Step 4, let d p be any data point in the k nearest neighbor data set k D (d p ) of the newly added data point d i in the UAV flight data set D, and the mutual nearest neighbor indication function s(d i ) of the data point d i is assigned a value according to the following formula:

[0013]

[0014] in the above formula denotes the class label of data point d i in the extended dataset D m ′ of UAV flight data,

[0015] Step 5, the working condition class C(d p ) of the new data point d p is determined according to the following formula:

[0016]

[0017] in the above formula, if the working condition is -1, it indicates that the new data point d p is a local outlier, and the process ends; if the working condition is not -1, the working condition of d p is C m , and the working condition of the new data point d p is output in the flowchart,

[0018] Step 6, if the number of working conditions to which the k nearest neighbor data set of the new data point d p in the UAV flight data set D belongs |lk D (d p )|≠1, randomly select a working condition n, where n∈lk D (d p ), add the new data point d p to the sub-data set D n to which the working condition n belongs, and form the extended data set D n ′,

[0019] Step 7, let data point d j be any one of the k nearest neighbor data set k D (d p ) of the new data point d p in the UAV flight data set D and the intersection of the UAV flight extended data set D n ′, and the mutual neighbor indicator function g(d j ) of data point d j is assigned according to the following formula:

[0020]

[0021] where, denotes the k nearest neighbor data set of data point d j in the UAV flight extended data set D n ′,

[0022] Step 8, the new data point d pNumber of mutual nearest neighbors | f(d p ) | :

[0023]

[0024] Step 9, if the newly added data point d p Number of mutual nearest neighbors | f(d p ) | = 0, then the working condition class label C(d p ) of the newly added data point d p ) = -2, mark the newly added data point d p as a global outlier, and the process ends;

[0025] If the newly added data point d p Number of mutual nearest neighbors 0 < |f(d<00000�6>)| < k, then jump to Step 10,

[0026] Step 10, assign a value to the position sequence number indicating function h(d p ) of the newly added data point d j relative to the data point d j , d p ) according to the following formula:

[0027]

[0028] In the above formula represents the position sequence number where the data point d n is located when the k nearest neighbors of the data point d j in the UAV flight extended data set D p ' are sorted in ascending order of distance.

[0029] [[ID=5ς]]Step 11, judge the working condition class label C(d p ) of the newly added data point d p ) according to the following formula:

[0030]

[0031] In the above formula |f(d p )| n represents the number of mutual nearest neighbors of the newly added data point d p in the working condition n data set. <00002з9>Compared with the prior art, the present invention has the following advantages:

[0033] The UAV working condition recognition seriously affects the accuracy of equipment fault prediction, fault diagnosis and remaining life assessment. In multi-task scenarios and noise interference, the UAV flight data distribution is uneven, showing variable density characteristics and there are outlier data. The above problems will seriously affect the accuracy of UAV flight data working condition recognition.

[0034] The present application can be applied to unmanned aerial vehicle variable density flight data. The working condition discrimination method based on mutual neighbor proposed by the present application can quickly determine the working condition attribution of newly added data in variable density data, and the working condition discrimination method is extended from uniformly distributed data to variable density data.

[0035] The outlying properties of data can be determined while identifying the working condition. The method of the present application can determine whether the newly added data is a global outlier and a local outlier in combination with the mutual neighbor rule of data, and gives the working condition discrimination algorithm the ability to identify outlier data, and the algorithm has stronger practicability and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a flowchart of the present application.

[0037] Figure 2 is a newly added data point d p Working condition class identification case 1.

[0038] Figure 3 is a newly added data point d p Working condition class identification case 2.

[0039] Figure 4 is a newly added data point d p Working condition class identification case 3. DETAILED DESCRIPTION

[0040] In order to facilitate those skilled in the art to understand and implement the present application, the present application will be further described in detail below in combination with examples, and it should be understood that the examples described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0041] Embodiment:

[0042] A working condition discrimination method for unmanned aerial vehicle variable density flight data based on mutual neighbor, comprising the following steps, as shown in Figure 1

[0043] Step 1, inputting unmanned aerial vehicle flight data newly added data point d p and unmanned aerial vehicle flight data set D. The unmanned aerial vehicle flight data set D is a known data set and contains l kinds (l=3 in this embodiment) of working condition sub-data sets D1, D2 and D3, wherein D1 represents a sub-data set in the unmanned aerial vehicle level flight working condition, D2 represents a sub-data set in the unmanned aerial vehicle climbing working condition, and D3 represents a sub-data set in the unmanned aerial vehicle diving working condition, and C1, C2 and C3 represent the unmanned aerial vehicle level flight, climbing and diving working condition classes respectively.

[0044] C() represents the working condition class label of a certain unmanned aerial vehicle flight data point, and k D ​() represents the set of k nearest neighbors of a given UAV flight data point in the UAV flight dataset D, denoted by lk. D () represents the operating conditions contained in the k nearest neighbor data sets of a given UAV flight data point in the UAV flight dataset D, denoted by |lk. D ()| represents the number of operating conditions contained in the k nearest neighbor data sets of a certain UAV flight data point in the UAV flight dataset D.

[0045] Step 2, combine the drone flight dataset D and the newly added data points d p Input the k-nearest neighbor algorithm (for ease of demonstrating the algorithm principle, the number of nearest neighbors k is set to 5) to obtain the newly added data point d. p In the drone flight dataset D, the k nearest neighbor data set k D (d p ) and newly added data point d p The k nearest neighbor datasets in the UAV flight dataset D contain lk of operating conditions. D (d p ).

[0046] Step 3, as follows Figure 2 As shown, if a new data point d is added... p The number of operating conditions to which the k nearest neighbor datasets belong in the UAV flight dataset D |lk D (d p In this embodiment, the newly added data point d is 1. p The set of k nearest neighbors k D (d p All data points in the data belong to the same operating condition C. m (In this embodiment, C1) then the new data point d will be added. p Add to the corresponding working condition subset D m In the case of m∈lk D (d p (In this embodiment, D1) forms the UAV flight extended dataset D. m (D1 in this embodiment), otherwise skip to step 6.

[0047] Step 4, let data point d i Add new data point d p In the drone flight dataset D, the k nearest neighbor data set k D (d p For any data point in ), the following formula represents the data point d. i Mutual nearest neighbor indicator function s(d i Assignment:

[0048]

[0049] In the above formula Represents data point d i In the extended dataset D of drone flights m The set of k nearest neighbors in ′.

[0050] Step 5, determine the newly added data point d according to the following formula. p Operating condition category C(d) p ):

[0051]

[0052] In the above formula, if the working condition judgment is -1, it means that a new data point d has been added. p The process ends when d is identified as a local outlier. p Working condition -1 corresponds to the newly added data point d in the flowchart. p The working condition; if the working condition judgment is not -1. If the working condition is not -1, d p The operating condition is Cm, which corresponds to the newly added data point d in the flowchart. p The operating conditions.

[0053] Step 6, if a new data point d is added p The number of operating conditions to which the k nearest neighbor datasets belong in the UAV flight dataset D |lk D (d p )|≠1, arbitrarily choose working condition n, where n∈lk D (d p ), will add new data point d p Add to the subset D to which working condition n belongs n In the middle, it forms the extended dataset D n ′.

[0054] Step 7, let data point d j Add new data point d p In the drone flight dataset D, the k nearest neighbor data set k D (d p ) and the extended dataset D of drone flights n Any data point in the intersection of ′ and ′ is represented by the following formula for data point d. j Mutual nearest neighbor indicator function g(d) j Assignment:

[0055]

[0056] in, Represents data point d j In the extended dataset D of drone flights n The set of k nearest neighbors in ′.

[0057] Step 8, calculate the new data point d p 's mutual neighbor number |f(d p )|

[0058]

[0059] Step 9, if the new data point d p 's mutual neighbor number |f(d p )| p =0, then the new data point d p 's working condition class label C(d p )=-2, that is, mark the new data point d p as a global outlier, and the process ends, at this time, the working condition of d p -2 corresponds to the working condition of outputting the new data point dp in the flow chart;

[0060] As shown in Figure 3 , if the new data point d p 's mutual neighbor number 0<|f(d p )|<k, then jump to step 10.

[0061] Step 10, assign the new data point d j 's position sequence number indication function h(d j ,d p ) according to the following formula:

[0062]

[0063] In the above formula, pos(d n ) represents the position sequence number of the data point d j in the k nearest neighbors of the data point d p in the unmanned aerial vehicle flight extended data set D p ', sorted in ascending order of distance,

[0064] Step 11, determine the new data point d p 's working condition class label C(d p ) according to the following formula:

[0065]

[0066] In the above formula, |f(d n )| p represents the mutual neighbor number of the new data point d p in the working condition n data set. As shown in Figure 4 , the k nearest neighbors of the new data point d p in the unmanned aerial vehicle flight data set D are scattered in C1, C2 and C3 working conditions, wherein the new data point d pThe number of mutual neighbors in the working condition C1 data set is 2 (data point d1 and data point d2), and the newly added data point d p The position sequence number indicating function values of the data point d1 and the data point d2 are 2 and 3 respectively, and the average indicating function value in the C1 working condition is (2+3) / 2=2.5. The average indicating function value in the C2 working condition is (5+5) / 2=5, and the average indicating function value in the C3 working condition is 0. Obviously, the working condition capable of making the average indicating function value take the minimum value is the working condition C1, and it is judged that the working condition class label of the newly added data point d p is C1.

[0067] It should be noted that the specific embodiments described in the present application are only examples illustrating the spirit of the present application. Those skilled in the art of the present application can make various modifications or supplements to the described specific embodiments or replace them with similar ways, but will not deviate from the spirit of the present application or exceed the scope defined by the appended claims.

Claims

1.A mutual neighbor-based method for identifying a working condition of variable-density flight data of a UAV, characterized in that, comprising the steps of Step 1, input new data point d to UAV flight data p and UAV flight data set D, UAV flight data set D is a known data set and contains l working condition sub-data sets D1~D l , C1~C l are working condition sub-data sets D1~D l corresponding working condition class labels, C() represents the working condition class label of a certain UAV flight data point, k D () represents the k nearest neighbor data set of a certain UAV flight data point in the UAV flight data set D, lk D () represents the working condition contained in the k nearest neighbor data set of a certain UAV flight data point in the UAV flight data set D, |lk D () represents the number of working conditions contained in the k nearest neighbor data set of a certain UAV flight data point in the UAV flight data set D, Step 2, obtain the new data point d from the UAV flight data set D and the new data point d p Input the k-neighbor algorithm to obtain the new data point d p The k nearest neighbor data set k in the UAV flight data set D D (d p ) and the new data point d p The working conditions lk contained in the k nearest neighbor data set in the UAV flight data set D D (d p ), Step 3, if a new data point d is added p The number of operating conditions to which the k nearest neighbor datasets belong in the UAV flight dataset D |lk D (d p )|=1, add a new data point d p The set of k nearest neighbors k D (d p All data points in the data belong to the same operating condition C. m Then the new data point d will be added. p Add to the corresponding working condition subset D m In the case of m∈lk D (d p This forms the extended dataset D of UAV flight. m Otherwise, proceed to step 6. Step 4, given data point d i For new data point d p In the k-nearest data set k D (d p ) of the k nearest data points in the UAV flight data set D, assign to the mutual nearest neighbor indicator function s(d i ) of data point d i the value: In the above formula representing data points d i In the drone flight extended dataset D m k-nearest data sets in the drone flight extended dataset D Step 5, determine the newly added data point d according to the following formula. p Operating condition category C(d) p ): In the above formula, if the working condition is judged as -1, it means that the new data point d is p a local outlier, and the process ends; if the working condition is not judged as -1, the working condition is not -1, d p is not a local outlier, and the working condition of d m is C p , which corresponds to the working condition of outputting the new data point d in the flowchart. Step 6, if new data point d p The number of working conditions to which the k nearest neighbor data set in the UAV flight data set D belongs |lk D (d p )≠1, any working condition n, where n∈lk D (d p ), add the new data point d p to the sub-data set D n to which the working condition n belongs, to form an extended data set D n ′, Step 7, let data point d j Add new data point d p In the drone flight dataset D, the k nearest neighbor data set k D (d p ) and the extended dataset D of drone flights n Any data point in the intersection of ′ and ′ is represented by the following formula for data point d. j Mutual nearest neighbor indicator function g(d) j Assignment: wherein, representing data points d j in the drone flight extended dataset D n k-nearest data sets in the drone flight extended dataset D Step 8, calculate the new data point d according to the following formula p The number of mutual neighbors of d p |f(d) | Step 9, if a new data point d is added p Number of mutual nearest neighbors | f(d) p If | = 0, then add a new data point d. p Operating condition category C(d) p ) = -2, adding new data point d p Mark as a global outlier, process ends; If a new data point d p The number of mutual neighbors 0 < |f(d p )| < k, then go to step 10, Step 10, assign a new data point d p according to the following formula j Position index function h(d j ,d p ) In the above formula This indicates the location of the extended dataset D for drone flights. n Data point d in ' j When the k nearest neighbors are sorted in ascending order of distance, the data point d p The position number is located at. Step 11, judge the new data point d according to the following formula p Condition class C(d p ) of the working condition | f(d p )| n represents the new data point d p The number of mutual neighbors in the data set at condition n.

Images