CN113946462A - Sensor system fault processing method and system for unmanned aerial vehicle cluster - Google Patents

Abstract

The invention discloses a method and a system for processing faults of a sensor system of an unmanned aerial vehicle cluster, wherein the method comprises the steps of obtaining the current output parameters of the sensor system of each unmanned aerial vehicle and generating training samples; carrying out standardization processing on the training samples, and carrying out dimensionality reduction processing based on principal component analysis; judging whether a sensor system of the unmanned aerial vehicle has a fault at present based on the training sample subjected to the dimensionality reduction treatment; if the fault exists, constructing a test sample by the current output parameter through a moving window method; carrying out standardization processing on the test sample, and carrying out dimensionality reduction processing based on principal component analysis; judging whether the sensor system of the unmanned aerial vehicle has a fault at present based on the test sample subjected to the dimensionality reduction treatment; if the fault exists, reconstructing the test sample, and searching a faulty sensor based on a reconstruction value; the reconstructed value is used as an output parameter of a sensor with a fault, so that soft maintenance is realized; the invention can realize the problems of fault detection and soft maintenance of the sensor system of the unmanned aerial vehicle cluster.

Description

Sensor system fault processing method and system for unmanned aerial vehicle cluster

Technical Field

The invention relates to a sensor system fault processing method and system for an unmanned aerial vehicle cluster, and belongs to the technical field of unmanned aerial vehicle clusters.

Background

The unmanned aerial vehicle cluster relates to the research object scope of complex system, and is widely applied to a plurality of fields such as agricultural irrigation, industrial production, battlefield monitoring and the like. How to improve the security, reliability and stability of the cluster is also a key problem which needs to be considered more and more in the current unmanned aerial vehicle cluster development process. The unmanned aerial vehicle individuals comprise core sensors such as a GPS (global positioning system), an IMU (inertial measurement unit), an altimeter, a magnetometer and an ADS (air data system), and the bottom layer architecture of the unmanned aerial vehicle individuals is shown in figure 1.

A cluster of drones can be understood as a complex multi-sensor system. In a multi-sensor system, in order to improve the reliability of the system operation, a hardware redundancy method is generally adopted to improve the reliability of the system. But in unmanned aerial vehicle system, for saving weight, can't hold redundant sensor. However, in the unmanned aerial vehicle cluster, the same sensors among all unmanned aerial vehicles often have strong correlation due to the fact that all unmanned aerial vehicles individually have similar moving modes. For large-scale equipment such as airplanes and the like, a method of regular maintenance is adopted to calibrate or maintain the sensors, and the sensors with obvious performance reduction or faults are replaced. However, the faults of each sensor and related circuits of the small unmanned aerial vehicle are not obvious, the faults are difficult to find through naked eyes, a large amount of manpower and financial resources are needed for regularly maintaining the system or replacing devices, and a large amount of unnecessary maintenance exists during maintenance and correction, so that the waste of a lot of resources is caused.

In order to solve the problem, the application provides a sensor system fault processing method and system of an unmanned aerial vehicle cluster.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, provides a sensor system fault processing method and system for an unmanned aerial vehicle cluster, and solves the problems that faults of sensors and related circuits of an unmanned aerial vehicle are not obvious, faults are difficult to find by naked eyes, a large amount of manpower and financial resources are needed for regularly maintaining the system or replacing devices, and a large amount of unnecessary maintenance exists during maintenance and correction, so that a lot of resources are wasted.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme:

in a first aspect, the present invention provides a method for handling a fault of a sensor system of an unmanned aerial vehicle cluster, including:

acquiring current output parameters of a sensor system of each unmanned aerial vehicle, and generating training samples;

carrying out standardization processing on the training samples, and carrying out dimensionality reduction processing based on principal component analysis;

judging whether a sensor system of the unmanned aerial vehicle has a fault at present based on the training sample subjected to the dimensionality reduction treatment;

if the fault exists, constructing a test sample by the current output parameter through a moving window method;

carrying out standardization processing on the test sample, and carrying out dimensionality reduction processing based on principal component analysis;

judging whether the sensor system of the unmanned aerial vehicle has a fault at present based on the test sample subjected to the dimensionality reduction treatment;

if the fault exists, reconstructing the test sample, and searching a faulty sensor based on a reconstruction value;

and taking the reconstructed value as an output parameter of the sensor with the fault, thereby realizing the soft maintenance of the sensor with the fault.

Preferably, the acquiring the current output parameters of the sensor system of each drone and generating the training samples includes:

acquiring output parameters of each sensor when a sensor system of the unmanned aerial vehicle works normally;

and taking the number of output parameters of the sensors as the number of rows and the number of the sensors as the number of columns to generate training samples.

Preferably, the normalizing the training samples and the dimensionality reduction based on the pivot analysis comprise:

carrying out standardization treatment on training samples:

wherein, X is a training sample,

for the normalized training sample, e (X) is the mean vector of the training sample X, denoted as e (X) [. mu. ]₁，μ₂，…，μ_n]，μ_nIs the mean value of the output parameter of the nth sensor, D_σFor training the sample X variance matrix, note

σ_nIs the standard deviation of the output parameter of the nth sensor, 1_mThe column vectors are column vectors with m-dimensional elements all being 1, and m is the number of output parameters of the sensor;

carrying out principal component analysis on the normalized training sample, and solving a correlation coefficient matrix R:

performing singular value analysis on the correlation coefficient matrix R:

R＝UΛU^T

wherein Λ is a characteristic matrix of R, and is denoted as Λ ═ diag (λ)_i，i＝1，2，…，n)，λ_iIs a characteristic value; u is a characteristic vector of R and is recorded as U ═ U₁，u₂，…，u_n]，u_nIs a column vector;

taking the front k (k < n) dimension of the feature vector U as a linear independent vector and recording as P ═ U₁，u₂，…，u_k]Forming a principal component load matrix; taking the k-n dimension of the feature vector U as a linear independent vector and recording the linear independent vector as

Forming a residual load matrix;

calculating the value of the main component quantity k based on the cumulative variance percentage principle;

and reducing the dimension of the normalized training sample based on the number k of the principal components.

Preferably, the calculating the value of the principal component number k based on the cumulative variance percentage principle includes:

calculating the cumulative contribution rate of the principal component load matrix:

wherein λ is_jIs the eigenvalue of the principal element load matrix, trace (R) is the trace of the principal element load matrix;

and solving the main component quantity k according to a preset CPV (k) value.

Preferably, the determining whether the sensor system of the unmanned aerial vehicle has a fault based on the training sample subjected to the dimension reduction processing includes:

converting the training sample after dimension reduction into a projection matrix C of a principal component space and a projection matrix of a residual error space

And obtaining a principal component space and a residual space, which are respectively recorded as

And

wherein the content of the first and second substances,

for training samples after dimensionality reduction, C ═ PP^T，

P is principal component load matrix，

As a residual loading matrix, I_n×nIs an n-dimensional identity matrix;

calculating SPE statistic of the training sample after dimension reduction in a residual error space based on the square prediction error:

judging SPE statistic and preset statistic threshold SPE_αThe size of the capsule is determined by the size of the capsule,

if SPE is more than or equal to SPE_αIf so, the sensor system is proved to have a fault;

if SPE < SPE_αThen the sensor system is proven to be free of faults.

Preferably, the preset threshold value SPE_aComprises the following steps:

wherein alpha is the check level, C_αIs a corresponding standard normal distribution of the signal,

is a parameter theta_iK is the number of principal components, and n is the number of sensors.

Preferably, the constructing the test sample by the moving window method with the current output parameter includes:

the current output parameter of the sensor system is compared with the adjacent s-1The output parameters are added to form a test data sample x^*；

x^*＝x_ij+x_i(j+1)+…+x_i(j+s)+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)

Where s denotes the window length, x_ijOutput parameter, Δ b, indicating that the ith sensor failed at time j_ijAs an output parameter x_ijA fault deviation of (2);

since the average value of the adjacent s-1 output parameters is close to 0 after the normalization process, the test data sample x is obtained^*＝x_ij+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)As a test sample.

Preferably, the sensor that reconstructs the test sample and finds a fault based on the reconstructed value includes:

and acquiring the optimal reconstruction value of the sensor by adopting a plurality of iterations:

reconstructing the test sample based on the principal component space:

the iterative expression is:

wherein x is_n(j) For the output parameter for which the nth sensor has a fault at time j,

C_jj-th column vector of projection matrix of pivot space, c_jjFor the projection matrix parameter, x, of the jth sensor at the jth time instant_jjAn output parameter indicating that the jth sensor failed at the jth time,

the reconstructed value of the output parameter of the ith sensor at the moment j; j is 1, 2, …, n,

and

respectively representing the output parameters x of the ith sensor_iA vector consisting of j-1 variables in the front and n-j variables in the back;

when converging to

Approach to

Then, a reconstruction formula is obtained:

calculating the difference value between the output parameter of each sensor in the sensor system and the reconstructed value of the output parameter, and acquiring the deviation value B at the fault moment_i(j)，

If it is a deviation value B_i(j) And if the deviation value is larger than the preset deviation value threshold value, the ith sensor fails.

In a second aspect, the present invention provides a sensor system fault handling system for an unmanned aerial vehicle cluster, which is characterized by comprising:

the data acquisition module is used for acquiring the current output parameters of the sensor system of the unmanned aerial vehicle and generating a training sample;

the first principal component analysis module is used for carrying out standardization processing on the training samples and carrying out dimensionality reduction processing based on principal component analysis;

the fault initial detection module is used for judging whether a sensor system of the unmanned aerial vehicle has faults at present based on the training samples subjected to the dimensionality reduction treatment;

the data framework module is used for constructing a test sample by the current output parameter through a moving window method if a fault exists;

the second principal component analysis module is used for carrying out standardized processing on the test sample and carrying out dimensionality reduction processing based on principal component analysis;

the fault rechecking module is used for judging whether the sensor system of the unmanned aerial vehicle has faults at present again based on the test sample subjected to the dimension reduction treatment;

the fault isolation module is used for reconstructing the test sample if a fault exists and searching a faulty sensor based on a reconstruction value;

and the soft maintenance module is used for taking the reconstructed value as an output parameter of the sensor with the fault so as to realize the soft maintenance of the sensor with the fault.

Compared with the prior art, the invention has the following beneficial effects:

according to the sensor system fault processing method and system for the unmanned aerial vehicle cluster, provided by the invention, the output values of the sensors can be subjected to online fault detection in the operation process of the sensor system of the unmanned aerial vehicle; after the system detects the trouble, can confirm the unmanned aerial vehicle's that breaks down fault sensor to carry out the reconfiguration to fault sensor's output value, realize the soft maintenance of trouble, thereby guarantee unmanned aerial vehicle's steady operation.

Drawings

Fig. 1 is a schematic diagram of the bottom architecture of an individual drone provided in the background of the invention;

fig. 2 is a flowchart of a method for processing a fault of a sensor system of an unmanned aerial vehicle cluster according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of ADS output results under different phase change conditions according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of SPE statistics of ADS under different phase transition conditions according to an embodiment of the present invention;

FIG. 5 is a schematic diagram illustrating an ADS output result of a CO-1 sensor having a power-off fault according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of SPE statistics of the ADS with the power-off failure of the CO-1 sensor according to the embodiment of the present invention;

FIG. 7 is a schematic diagram of SPE statistics of a moving window method for a power failure of a CO-1 sensor according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of the result of the isolation of a CO-1 sensor in a power-off fault according to an embodiment of the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.

The first embodiment is as follows:

as shown in fig. 2, the present embodiment provides a method for processing a fault of a sensor system of an unmanned aerial vehicle cluster, including the following steps:

step 1, obtaining current output parameters of a sensor system of each unmanned aerial vehicle, and generating training samples;

acquiring output parameters of each sensor when a sensor system of the unmanned aerial vehicle works normally;

and taking the number of output parameters of the sensors as the number of rows and the number of the sensors as the number of columns to generate training samples.

Step 2, carrying out standardization processing on the training samples, and carrying out dimensionality reduction processing based on principal component analysis;

carrying out standardization treatment on training samples:

wherein, X is a training sample,

for the normalized training sample, e (X) is the mean vector of the training sample X, denoted as e (X) [. mu. ]₁，μ₂，…，μ_n]，μ_nIs the mean value of the output parameter of the nth sensor, D_σFor training the sample X variance matrix, note

σ_nIs the standard deviation of the output parameter of the nth sensor, 1_mThe column vectors are column vectors with m-dimensional elements all being 1, and m is the number of output parameters of the sensor;

carrying out principal component analysis on the normalized training sample, and solving a correlation coefficient matrix R:

performing singular value analysis on the correlation coefficient matrix R:

R＝UΛU^T

wherein Λ is a characteristic matrix of R, and is denoted as Λ ═ diag (λ)_i，i＝1，2，…，n)，λ_iIs a characteristic value; u is a characteristic vector of R and is recorded as U ═ U₁，u₂，…，u_n]，u_nIs a column vector;

taking the front k (k < n) dimension of the feature vector U as a linear independent vector and recording as P ═ U₁，u₂，…，u_k]Forming a principal component load matrix; taking the k-n dimension of the feature vector U as a linear independent vector and recording the linear independent vector as

Forming a residual load matrix;

calculating the value of the main component quantity k based on the cumulative variance percentage principle;

and reducing the dimension of the normalized training sample based on the number k of the principal components.

The method for solving the value of the principal component number k based on the cumulative variance percentage principle comprises the following steps:

calculating the cumulative contribution rate of the principal component load matrix:

wherein λ is_jAs principal component load matrixEigenvalue, trace (r), is the trace of the principal element load matrix;

and solving the main component quantity k according to a preset CPV (k) value (generally 95%).

Step 3, judging whether a sensor system of the unmanned aerial vehicle has a fault at present based on the training sample subjected to the dimensionality reduction treatment;

converting the training sample after dimension reduction into a projection matrix C of a principal component space and a projection matrix of a residual error space

And obtaining a principal component space and a residual space, which are respectively recorded as

And

wherein the content of the first and second substances,

for training samples after dimensionality reduction, C ═ PP^T，

P is a principal component load matrix,

as a residual loading matrix, I_n×nIs an n-dimensional identity matrix;

calculating SPE statistic of the training sample after dimension reduction in a residual error space based on the square prediction error:

judging SPE statistic and preset statistic threshold SPE_αThe size of the capsule is determined by the size of the capsule,

if SPE is more than or equal to SPE_αIf so, the sensor system is proved to have a fault;

if SPE < SPE_αThen the sensor system is proven to be free of faults.

Wherein, a preset threshold SPE_αComprises the following steps:

wherein alpha is the check level, C_αIs a corresponding standard normal distribution of the signal,

is a parameter theta_iK is the number of principal components, and n is the number of sensors.

Step 4, if a fault exists, constructing a test sample by the current output parameter through a moving window method;

adding the current output parameter of the sensor system with s-1 adjacent output parameters to form a test data sample x^*；

x^*＝x_ij+x_i(j+1)+…+x_i(j+s)+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)

Where s denotes the window length, x_ijOutput parameter, Δ b, indicating that the ith sensor failed at time j_ijAs an output parameter x_ijA fault deviation of (2);

because adjacent s-1 output parameters are normalizedTreatment, with an average approaching 0, samples x of test data^*＝x_ij+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)As a test sample.

Step 5, carrying out standardization processing on the test sample, and carrying out dimensionality reduction processing based on principal component analysis; the process is the same principle as step 2.

Step 6, judging whether the sensor system of the unmanned aerial vehicle has faults at present again based on the test sample subjected to the dimension reduction treatment; the process is the same principle as step 3.

Step 7, if a fault exists, reconstructing the test sample, and searching a faulty sensor based on a reconstructed value;

and acquiring the optimal reconstruction value of the sensor by adopting a plurality of iterations:

reconstructing the test sample based on the principal component space:

the iterative expression is:

wherein x is_n(j) For the output parameter for which the nth sensor has a fault at time j,

C_jj-th column vector of projection matrix of pivot space, c_jjFor the projection matrix parameter, x, of the jth sensor at the jth time instant_jjAn output parameter indicating that the jth sensor failed at the jth time,

the reconstructed value of the output parameter of the ith sensor at the moment j; j is 1, 2, …, n,

and 1

Respectively representing the output parameters x of the ith sensor_iA vector consisting of j-1 variables in the front and n-j variables in the back;

when converging to

Approach to

Then, a reconstruction formula is obtained:

calculating the difference value between the output parameter of each sensor in the sensor system and the reconstructed value of the output parameter, and acquiring the deviation value B at the fault moment_i(j)，

If it is a deviation value B_i(j) And if the deviation value is larger than the preset deviation value threshold value, the ith sensor fails.

And 8, taking the reconstructed value as an output parameter of the sensor with the fault, thereby realizing the soft maintenance of the sensor with the fault.

The validity verification of the processing method provided by the embodiment:

(1) abnormal state detection of normal signals

Assuming that the unmanned aerial vehicle cluster normally works, all sensors work in a fault-free state. In order to verify the effectiveness of the method, three unmanned aerial vehicles are adopted to output ADS results under the conditions of steady-state signals and different phase changes, response values output by the sensors under different phase changes are respectively collected, and a test sample is constructed as shown in FIG. 3. Before the phase change does not occur, the output value can be regarded as a steady-state signal. And then, analyzing the signals by using a fault detection model of a principal component analysis method, and judging whether the signals can effectively distinguish normal signals from fault signals under the non-fault condition.

Based on the principle of fault detection of the principal component analysis method, the cumulative percentage of variance of the first principal component is 58.6 percent and is less than the set threshold value of 95 percent, the cumulative percentage of variance of the first two principal components is 96.97 percent and exceeds the set threshold value, so that the number of the principal components is selected to be 2, and the reason that certain correlation exists among the sensors is that the same gas sensors exist in the system. The threshold for the SPE statistic can be calculated when the confidence level of the gaussian distribution is 95% in the residual space. Analysis of the test sample shown in FIG. 3 allows the calculation of the SPE statistics, as shown in FIG. 4. Therefore, whether the signal is a stable signal or a normal sudden change signal, the normal signal can be accurately identified by applying the pivot analysis method, and the false alarm condition cannot occur.

(2) Abnormal state detection in case of fault

When a certain sensor in the system fails, the system is provided with whether the system fails or not, and meanwhile, the isolation and recovery of the failure are required to be effectively realized, namely, the position of the system which fails and the optimal estimation value are provided. Taking the power failure fault of the CO-1 sensor in the ADS as an example, the method carries out the simulation of fault detection, isolation and recovery, and verifies the feasibility of the method used in the text.

As shown in FIG. 5, taking the power failure as an example, the sensor CO-1 has a power failure at the 100 th s, and its measurement value rapidly drops from 2.5V to 0V. At this time, as shown in fig. 6, the SPE statistic rapidly rises and exceeds the set threshold SPE_αThe existence of one or more faults in the sensor system at this time is proven. In order to further confirm the details of the fault, a moving window principal component analysis method is adopted for further fault condition analysis. The signal in fig. 5 is further decomposed, and feature extraction is performed by applying a moving window, and the obtained result is shown in fig. 7. Therefore, the fault detection result obtained by applying the method is still the fault in the cluster.

As shown in fig. 8, in order to determine the fault location of the sensor, a PCA reconstruction method is applied to reconstruct data, and a system deviation value at the fault time is obtained. At the moment, the deviation value of the sensor CO-1 is far larger than an expected value, which proves that the sensor CO-1 has serious faults at the moment, and the deviation values of other sensors are all in a normal error allowable range, so that the sensors are all in a normal working state. The specific offset values in the offset vector at this time are shown in table 1 below.

TABLE 1 deviation values of sensors of a sensor system in the event of a CO-1 outage

Table 1 Bias value of each sensor in sensor system under outage fault of CO-1

Example two:

the embodiment provides an unmanned aerial vehicle's sensor system fault handling system, includes:

the data acquisition module is used for acquiring the current output parameters of the sensor system of the unmanned aerial vehicle and generating a training sample;

the first principal component analysis module is used for carrying out standardization processing on the training samples and carrying out dimensionality reduction processing based on principal component analysis;

the fault initial detection module is used for judging whether a sensor system of the unmanned aerial vehicle has faults at present based on the training samples subjected to the dimensionality reduction treatment;

the data framework module is used for constructing a test sample by the current output parameter through a moving window method if a fault exists;

the second principal component analysis module is used for carrying out standardized processing on the test sample and carrying out dimensionality reduction processing based on principal component analysis;

the fault rechecking module is used for judging whether the sensor system of the unmanned aerial vehicle has faults at present again based on the test sample subjected to the dimension reduction treatment;

the fault isolation module is used for reconstructing the test sample if a fault exists and searching a faulty sensor based on a reconstruction value;

and the soft maintenance module is used for taking the reconstructed value as an output parameter of the sensor with the fault so as to realize the soft maintenance of the sensor with the fault.

The invention provides a sensor system fault processing method and system of an unmanned aerial vehicle cluster, aiming at solving the problem that the unmanned aerial vehicle cluster is difficult to effectively detect the self state in complex environment or electromagnetic interference and other scenes or in burst states, and the method and system can realize three core functions of the unmanned aerial vehicle cluster: 1. the output values of all sensors of the unmanned aerial vehicle can be subjected to online fault detection in the system operation process; 2. when the system detects a fault, it can be determined which sensor(s) of which drone(s) failed; 3. and reconstructing the information of the fault sensor, and giving the normal output value of the fault sensor to realize the soft recovery of the fault.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (9)

1. A sensor system fault processing method of an unmanned aerial vehicle cluster is characterized by comprising the following steps:

acquiring current output parameters of a sensor system of each unmanned aerial vehicle, and generating training samples;

carrying out standardization processing on the training samples, and carrying out dimensionality reduction processing based on principal component analysis;

judging whether a sensor system of the unmanned aerial vehicle has a fault at present based on the training sample subjected to the dimensionality reduction treatment;

if the fault exists, constructing a test sample by the current output parameter through a moving window method;

carrying out standardization processing on the test sample, and carrying out dimensionality reduction processing based on principal component analysis;

judging whether the sensor system of the unmanned aerial vehicle has a fault at present based on the test sample subjected to the dimensionality reduction treatment;

if the fault exists, reconstructing the test sample, and searching a faulty sensor based on a reconstruction value;

and taking the reconstructed value as an output parameter of the sensor with the fault, thereby realizing the soft maintenance of the sensor with the fault.

2. The method of claim 1, wherein the obtaining current output parameters of the sensor system of each drone and generating training samples comprises:

acquiring output parameters of each sensor when a sensor system of the unmanned aerial vehicle works normally;

and taking the number of output parameters of the sensors as the number of rows and the number of the sensors as the number of columns to generate training samples.

3. The method of claim 1, wherein the normalizing the training samples and the dimensionality reduction based on the principal component analysis comprise:

carrying out standardization treatment on training samples:

wherein, X is a training sample,

for the normalized training sample, e (X) is the mean vector of the training sample X, denoted as e (X) [. mu. ]₁，μ₂，…，μ_n]，μ_nIs the mean value of the output parameter of the nth sensor, D_σFor training the sample X variance matrix, note

σ_nIs the output of the nth sensorStandard deviation of parameters, 1_mThe column vectors are column vectors with m-dimensional elements all being 1, and m is the number of output parameters of the sensor;

carrying out principal component analysis on the normalized training sample, and solving a correlation coefficient matrix R:

performing singular value analysis on the correlation coefficient matrix R:

R＝UΛU^T

wherein Λ is a characteristic matrix of R, and is denoted as Λ ═ diag (λ)_i，i＝1，2，…，n)，λ_iIs a characteristic value; u is a characteristic vector of R and is recorded as U ═ U₁，u₂，…，u_n]，u_nIs a column vector;

taking the front k (k < n) dimension of the feature vector U as a linear independent vector and recording as P ═ U₁，u₂，…，u_k]Forming a principal component load matrix; taking the k-n dimension of the feature vector U as a linear independent vector and recording the linear independent vector as

Forming a residual load matrix;

calculating the value of the main component quantity k based on the cumulative variance percentage principle;

and reducing the dimension of the normalized training sample based on the number k of the principal components.

4. The method of claim 3, wherein the calculating the value of the principal component number k based on the cumulative variance percentage principle comprises:

calculating the cumulative contribution rate of the principal component load matrix:

wherein λ is_jIs the eigenvalue of the principal element load matrix, trace (R) is the trace of the principal element load matrix;

and solving the main component quantity k according to a preset CPV (k) value.

5. The method of claim 1, wherein the determining whether the sensor system of the drone has a fault based on the training samples processed in the dimensionality reduction process comprises:

converting the training sample after dimension reduction into a projection matrix C of a principal component space and a projection matrix of a residual error space

And obtaining a principal component space and a residual space, which are respectively recorded as

And

wherein the content of the first and second substances,

for training samples after dimensionality reduction, C ═ PP^T，

P is a principal component load matrix,

as a residual loading matrix, I_n×nIs an n-dimensional identity matrix;

calculating SPE statistic of the training sample after dimension reduction in a residual error space based on the square prediction error:

judging SPE statistic and preset statistic threshold SPE_αThe size of the capsule is determined by the size of the capsule,

if SPE is more than or equal to SPE_αIf so, the sensor system is proved to have a fault;

if SPE < SPE_αThen the sensor system is proven to be free of faults.

6. The method as claimed in claim 5, wherein the predetermined threshold SPE is a threshold_αComprises the following steps:

wherein alpha is the check level, C_αIs a corresponding standard normal distribution of the signal,

is a parameter theta_iK is the number of principal components, and n is the number of sensors.

7. The method of claim 1, wherein the constructing test samples from the current output parameters by a moving window method comprises:

adding the current output parameter of the sensor system with s-1 adjacent output parameters to form a test data sample x^*；

x^*＝x_ij+x_i(j+1)+…+x_i(j+s)+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)

Where s denotes the window length, x_ijOutput parameter, Δ b, indicating that the ith sensor failed at time j_ijAs an output parameter x_ijA fault deviation of (2);

since the average value of the adjacent s-1 output parameters is close to 0 after the normalization process, the test data sample x is obtained^*＝x_ij+Δb_ij+Δb_i(j+1)+…+Δb_i(j+s)As a test sample.

8. The method of claim 1, wherein reconstructing the test sample and finding the faulty sensor based on the reconstructed value comprises:

and acquiring the optimal reconstruction value of the sensor by adopting a plurality of iterations:

reconstructing the test sample based on the principal component space:

the iterative expression is:

wherein x is_n(j) For the output parameter for which the nth sensor has a fault at time j,

C_jj-th of projection matrix as principal component spaceColumn vector, c_jjFor the projection matrix parameter, x, of the jth sensor at the jth time instant_jjAn output parameter indicating that the jth sensor failed at the jth time,

the reconstructed value of the output parameter of the ith sensor at the moment j; j is 1, 2, …, n,

and

respectively representing the output parameters x of the ith sensor_iA vector consisting of j-1 variables in the front and n-j variables in the back;

when converging to

Approach to

Then, a reconstruction formula is obtained:

calculating the difference value between the output parameter of each sensor in the sensor system and the reconstructed value of the output parameter, and acquiring the deviation value B at the fault moment_i(j)，

If it is a deviation value B_i(t) is greater than the preset deviation value threshold, the ith sensor fails.

9. A sensor system fault handling system of unmanned aerial vehicle cluster, its characterized in that includes:

the data acquisition module is used for acquiring the current output parameters of the sensor system of the unmanned aerial vehicle and generating a training sample;

the first principal component analysis module is used for carrying out standardization processing on the training samples and carrying out dimensionality reduction processing based on principal component analysis;

the fault initial detection module is used for judging whether a sensor system of the unmanned aerial vehicle has faults at present based on the training samples subjected to the dimensionality reduction treatment;

the data framework module is used for constructing a test sample by the current output parameter through a moving window method if a fault exists;

the second principal component analysis module is used for carrying out standardized processing on the test sample and carrying out dimensionality reduction processing based on principal component analysis;

the fault rechecking module is used for judging whether the sensor system of the unmanned aerial vehicle has faults at present again based on the test sample subjected to the dimension reduction treatment;

the fault isolation module is used for reconstructing the test sample if a fault exists and searching a faulty sensor based on a reconstruction value;

and the soft maintenance module is used for taking the reconstructed value as an output parameter of the sensor with the fault so as to realize the soft maintenance of the sensor with the fault.

Images