A sensor system fault processing method and system of a drone cluster

Abstract

The application discloses a kind of unmanned aerial vehicle cluster's sensor system fault processing method and system, its method includes obtaining the current output parameter of the sensor system of each unmanned aerial vehicle, and generates training sample;Training sample is standardized and is processed, and dimension reduction is processed based on principal component analysis;Whether the sensor system of unmanned aerial vehicle currently exists fault based on the training sample of dimension reduction processing;If there is fault, current output parameter is constructed test sample by mobile window method;Test sample is standardized and is processed, and dimension reduction is processed based on principal component analysis;Whether the sensor system of unmanned aerial vehicle currently exists fault based on the test sample of dimension reduction processing again;If there is fault, test sample is reconstructed, and the sensor of fault is found based on reconstruction value;Reconstruction value is used as the output parameter of the sensor of fault, to realize soft maintenance;The application can realize the sensor system fault detection and soft maintenance problem of unmanned aerial vehicle cluster.

Description

TECHNICAL FIELD

[0001] The present application relates to a sensor system fault processing method and system of a UAV cluster, and belongs to the technical field of a UAV cluster. BACKGROUND

[0002] The UAV cluster relates to the research object category of a complex system, and is widely used in agricultural irrigation, industrial production, battlefield monitoring and many other fields. How to improve the safety, reliability and stability of the cluster is also a key problem that needs to be considered in the development process of the current UAV cluster. The UAV individual includes core sensors such as GPS, IMU, altimeter, magnetometer and ADS (air data system), and the bottom structure is as shown in FIG. Figure 1

[0003] The UAV cluster can be understood as a complex multi-sensor system. In a multi-sensor system, in order to improve the reliability of the system, a hardware redundancy method is usually used to improve the reliability of the system. However, in the UAV system, due to the need to save weight, redundant sensors cannot be accommodated. However, in the UAV cluster, each UAV individual has a similar movement mode, and the same sensors between each UAV usually have a strong correlation. For large equipment such as aircraft, a regular maintenance method is used to calibrate or maintain the sensors, and to replace sensors with obvious performance degradation or faults. However, the faults of each sensor and related circuit of the small UAV are not obvious, and it is difficult to find faults by the naked eye. Regular maintenance systems or replacement of devices also require a large amount of manpower and financial resources, and there is a lot of unnecessary maintenance during maintenance correction, resulting in a waste of many resources.

[0004] In order to solve the above problems, the present application provides a sensor system fault processing method and system of a UAV cluster. SUMMARY

[0005] The purpose of the present application is to overcome the deficiencies in the prior art, provide a sensor system fault processing method and system of a UAV cluster, and solve the problem that the faults of each sensor and related circuit of the UAV are not obvious, it is difficult to find faults by the naked eye, regular maintenance systems or replacement of devices also require a large amount of manpower and financial resources, and there is a lot of unnecessary maintenance during maintenance correction, resulting in a waste of many resources.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0007] In a first aspect, the present application provides a sensor system fault processing method of a UAV cluster, comprising:

[0008] Obtaining the current output parameters of the sensor system of each UAV, and generating a training sample;

[0009] ​The training samples are standardized and then dimensionality is reduced based on principal component analysis.

[0010] Determine whether the drone's sensor system is currently faulty based on the training samples processed by dimensionality reduction;

[0011] If a fault exists, the current output parameters will be used to construct a test sample using the moving window method;

[0012] The test samples were standardized and then dimensionality was reduced based on principal component analysis.

[0013] Based on the dimensionality reduction of the test samples, we can further determine whether the drone's sensor system is currently faulty;

[0014] If a fault is found, the test sample is reconstructed, and the faulty sensor is located based on the reconstructed values.

[0015] By using the reconstructed value as the output parameter of the faulty sensor, soft repair of the faulty sensor can be achieved.

[0016] Preferably, the step of obtaining the current output parameters of the sensor system of each UAV and generating training samples includes:

[0017] Acquire the output parameters of each sensor when the drone's sensor system is working normally;

[0018] Training samples are generated using the number of output parameters of the sensors as the number of rows and the number of sensors as the number of columns.

[0019] Preferably, the standardization of training samples and the dimensionality reduction based on principal component analysis include:

[0020] Standardize the training samples:

[0021]

[0022] Where X represents the training samples. For the standardized training samples, E(X) is the mean vector of the training samples X, denoted as E(X) = [μ1, μ2, ..., μ...]. n ], μ n Let D be the mean of the output parameters of the nth sensor. σ Let X be the variance matrix of the training samples, denoted as σ n Let be the standard deviation of the output parameters of the nth sensor, 1 m It is an m-dimensional column vector with all elements equal to 1, where m is the number of output parameters of the sensor;

[0023] Principal component analysis was performed on the standardized training samples to obtain their correlation coefficient matrix R:

[0024]

[0025] singular value analysis is performed on the correlation coefficient matrix R:

[0026] R = UΛU T

[0027] wherein, Λ is the eigenmatrix of R, denoted as Λ = diag(λ i , i = 1, 2, …, n), λ i is the eigenvalue; U is the eigenvector of R, denoted as U = [u1, u2, …, u n ], u n is a column vector;

[0028] The first k dimensions of the eigenvector U are taken as linearly independent vectors, denoted as P = [u1, u2, …, u k ], to form a principal component loading matrix; the k+1 dimensions to n dimensions of the eigenvector U are taken as linearly independent vectors, denoted as to form a residual loading matrix;

[0029] The value of the principal component number k is obtained based on the cumulative variance percentage principle;

[0030] The training samples after standardization are reduced in dimension based on the principal component number k.

[0031] Preferably, the value of the principal component number k is obtained based on the cumulative variance percentage principle includes:

[0032] The cumulative contribution rate of the principal component loading matrix is calculated:

[0033]

[0034] wherein, λ j is the eigenvalue of the principal component loading matrix, and trace(R) is the trace of the principal component loading matrix;

[0035] The principal component number k is solved according to the preset CPV(k) value.

[0036] Preferably, the determination of whether the sensor system of the unmanned aerial vehicle currently exists a fault based on the training samples after the dimension reduction processing includes:

[0037] The training samples after the dimension reduction are converted into the projection matrix C of the principal component space and the projection matrix of the residual space, and the principal component space and the residual space are obtained, denoted as and

[0038] wherein, For the dimensionality-reduced training samples, C = PP T , P is the principal element load matrix. Let I be the residual loading matrix. n×n It is an n-dimensional identity matrix;

[0039] The SPE statistic, calculated based on the squared prediction error, shows the variation of the dimensionality-reduced training samples within the residual space.

[0040]

[0041] Determine if the SPE statistic matches the preset SPE threshold. α size,

[0042] If SPE≥SPE α This proves that the sensor system is faulty;

[0043] If SPE <SPE α This proves that the sensor system is not faulty.

[0044] Preferably, the preset threshold SPE α for:

[0045]

[0046] Where α is the validation level, C α For the corresponding standard normal distribution, For parameter θ i The j-th feature is given by k, the number of principal components is given by n, and the number of sensors is given by n.

[0047] Preferably, constructing test samples using the moving window method based on the current output parameters includes:

[0048] The current output parameter of the sensor system is added to the adjacent s-1 output parameters to form the test data sample x. * ;

[0049] x * =x ij +x i(j+1) +…+x i(j+s-1) +Δb ij +Δb i(j+1) +…+Δb i(j+s-1)

[0050] Where s represents the window length, x ij Δb represents the output parameter where the i-th sensor fails at time j. ij For the output parameter x ij Fault deviation;

[0051] Since the average value of the adjacent s-1 output parameters is close to 0 after standardization, the test data sample x * =x ij +Δb ij +Δb i(j+1) +…+Δb i(j+s-1) As a test sample.

[0052] Preferably, the sensor for reconstructing the test sample and locating the fault based on the reconstructed values ​​includes:

[0053] The optimal reconstruction value of the sensor is obtained by using multiple iterations:

[0054] Reconstructing the test samples based on the principal component space:

[0055]

[0056] The iterative expression is:

[0057]

[0058] Where, x n (j) represents the output parameter of the nth sensor at time j when the sensor is faulty. C j c is the j-th column vector of the projection matrix of the principal element space. jj Let x be the projection matrix parameter of the j-th sensor at time j. jj This represents the output parameter indicating that the j-th sensor malfunctions at time j. Let be the reconstructed value of the output parameter of the i-th sensor at time j; j = 1, 2, ..., n. Let x be the output parameter of the i-th sensor. i The vector consisting of the first j-1 variables, Let x be the output parameter of the i-th sensor. i A vector consisting of the last nj variables;

[0059] when Approaching Upon convergence, the reconstruction formula is obtained:

[0060]

[0061] Calculate the difference between the output parameters of each sensor in the sensor system and the reconstructed values ​​of the output parameters, and obtain the deviation value B at the time of the fault. i (j),

[0062] If the deviation value B iIf (j) is greater than the preset deviation threshold, then the i-th sensor is faulty.

[0063] Secondly, the present invention provides a sensor system fault handling system for a drone swarm, comprising:

[0064] The data acquisition module is used to acquire the current output parameters of the UAV's sensor system and generate training samples;

[0065] The first principal component analysis module is used to standardize the training samples and perform dimensionality reduction based on principal component analysis.

[0066] The initial fault detection module is used to determine whether there is a fault in the drone's sensor system based on the training samples processed by dimensionality reduction;

[0067] The data architecture module is used to construct test samples from the current output parameters using the moving window method if a fault exists.

[0068] The second principal component analysis module is used to standardize the test samples and perform dimensionality reduction based on principal component analysis.

[0069] The fault re-examination module is used to re-determine whether the drone's sensor system is currently faulty based on the test samples after dimensionality reduction processing;

[0070] The fault isolation module is used to reconstruct the test sample if a fault exists, and to find the faulty sensor based on the reconstructed values.

[0071] The soft repair module is used to use the reconstructed value as the output parameter of the faulty sensor, thereby enabling soft repair of the faulty sensor.

[0072] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0073] The present invention provides a sensor system fault handling method and system for a drone swarm, which can perform online fault detection on the output values ​​of each sensor during the operation of the drone's sensor system; when the system detects a fault, it can identify the faulty sensor of the drone that has malfunctioned and reconstruct the output value of the faulty sensor to achieve soft repair of the fault, thereby ensuring the stable operation of the drone. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of the underlying architecture of an individual drone provided in the background technology of this invention;

[0075] Figure 2 This is a flowchart of a sensor system fault handling method for a drone swarm provided in an embodiment of the present invention;

[0076] Figure 3This is a schematic diagram of ADS output results under different phase transition conditions provided in the embodiments of the present invention;

[0077] Figure 4 This is a schematic diagram of the SPE statistics of ADS under different phase transition conditions provided in the embodiments of the present invention;

[0078] Figure 5 This is a schematic diagram of the ADS output results when the CO-1 sensor experiences a power outage fault, as provided in an embodiment of the present invention.

[0079] Figure 6 This is a schematic diagram of the SPE statistics of the ADS when the CO-1 sensor experiences a power failure, as provided in this embodiment of the invention.

[0080] Figure 7 This is a schematic diagram of the SPE statistics of the moving window method for a CO-1 sensor power failure provided in an embodiment of the present invention;

[0081] Figure 8 This is a schematic diagram of the isolation result when the CO-1 sensor experiences a power outage fault, as provided in an embodiment of the present invention. Detailed Implementation

[0082] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0083] Example 1:

[0084] like Figure 2 As shown, this embodiment provides a method for handling sensor system faults in a drone swarm, including the following steps:

[0085] Step 1: Obtain the current output parameters of the sensor system for each drone and generate training samples;

[0086] Acquire the output parameters of each sensor when the drone's sensor system is working normally;

[0087] Training samples are generated using the number of output parameters of the sensors as the number of rows and the number of sensors as the number of columns.

[0088] Step 2: Standardize the training samples and perform dimensionality reduction based on principal component analysis;

[0089] Standardize the training samples:

[0090]

[0091] Where X represents the training samples. The training samples after standardization, E(X) is the mean vector of the training sample X, denoted as E(X) = [μ1, μ2, …, μ n , μ n is the mean of the output parameters of the nth sensor, D σ is the variance matrix of the training sample X, denoted as σ n is the standard deviation of the output parameters of the nth sensor, 1 m is a column vector with all m - dimensional elements being 1, and m is the number of output parameters of the sensor;

[0092] Perform principal component analysis on the standardized training samples to obtain its correlation coefficient matrix R:

[0093]

[0094] Perform singular value analysis on the correlation coefficient matrix R:

[0095] R = UΛU T

[0096] where Λ is the eigenmatrix of R, denoted as Λ = diag(λ i , i = 1, 2, …, n), λ i is the eigenvalue; U is the eigenvector of R, denoted as U = [u1, u2, …, u n , u n is a column vector;

[0097] Take the first k dimensions of the eigenvector U as linearly independent vectors, k < n, denoted as P = [u1, u2, …, u k , to form the principal component loading matrix; Take the k + 1 - dimensional to n - dimensional of the eigenvector U as linearly independent vectors, denoted as to form the residual loading matrix;

[0098] Obtain the value of the number of principal components k based on the cumulative variance percentage principle;

[0099] Reduce the dimension of the standardized training samples based on the number of principal components k.

[0100] Among them, obtaining the value of the number of principal components k based on the cumulative variance percentage principle includes:

[0101] Calculate the cumulative contribution rate of the principal component loading matrix:

[0102]

[0103] where λ j is the eigenvalue of the principal component loading matrix, trace(R) is the trace of the principal component loading matrix;

[0104] The number of principal components k is calculated based on the preset CPV(k) value (usually taken as 95%).

[0105] Step 3: Determine whether the drone's sensor system is currently faulty based on the dimensionality-reduced training samples;

[0106] The projection matrix C of the dimensionality-reduced training samples to the principal space and the projection matrix of the residual space are transformed. In the process, the principal space and residual space are obtained, denoted as […]. and

[0107] in, For the dimensionality-reduced training samples, C = PP T , P is the principal element load matrix. Let I be the residual loading matrix. n×n It is an n-dimensional identity matrix;

[0108] The SPE statistic, calculated based on the squared prediction error, shows the variation of the dimensionality-reduced training samples within the residual space.

[0109]

[0110] Determine if the SPE statistic matches the preset SPE threshold. α size,

[0111] If SPE≥SPE α This proves that the sensor system is faulty;

[0112] If SPE <SPE α This proves that the sensor system is not faulty.

[0113] Among them, the preset threshold SPE α for:

[0114]

[0115]

[0116] Where α is the validation level, C α For the corresponding standard normal distribution, For parameter θ i The j-th feature is given by k, the number of principal components is given by n, and the number of sensors is given by n.

[0117] Step 4: If a fault exists, construct a test sample using the moving window method with the current output parameters;

[0118] The current output parameter of the sensor system is added to the adjacent s-1 output parameters to form the test data sample x. * ;

[0119] x * =x ij +x i(j+1) +…+x i(j+s-1) +Δb ij +Δb i(j+1) +…+Δb i(j+s-1)

[0120] Where s represents the window length, x ij Δb represents the output parameter where the i-th sensor fails at time j. ij For the output parameter x ij Fault deviation;

[0121] Since the average value of the adjacent s-1 output parameters is close to 0 after standardization, the test data sample x * =x ij +Δb ij +Δb i(j+1) +…+Δb i(j+s-1) As a test sample.

[0122] Step 5: Standardize the test samples and perform dimensionality reduction based on principal component analysis; the process is the same as in Step 2.

[0123] Step 6: Based on the dimensionality reduction test samples, determine again whether the drone's sensor system is currently faulty; the process is the same as in Step 3.

[0124] Step 7: If a fault exists, reconstruct the test sample and locate the faulty sensor based on the reconstructed values;

[0125] The optimal reconstruction value of the sensor is obtained by using multiple iterations:

[0126] Reconstructing the test samples based on the principal component space:

[0127]

[0128] The iterative expression is:

[0129]

[0130] Where, x n (j) represents the output parameter of the nth sensor at time j when the sensor is faulty. C j c is the j-th column vector of the projection matrix of the principal element space. jjLet x be the projection matrix parameter of the j-th sensor at time j. jj This represents the output parameter indicating that the j-th sensor malfunctions at time j. Let be the reconstructed value of the output parameter of the i-th sensor at time j; j = 1, 2, ..., n. Let x be the output parameter of the i-th sensor. i The vector consisting of the first j-1 variables, Let x be the output parameter of the i-th sensor. i A vector consisting of the last nj variables;

[0131] When convergence to Approaching Then, the reconstruction formula is obtained:

[0132]

[0133] Calculate the difference between the output parameters of each sensor in the sensor system and the reconstructed values ​​of the output parameters, and obtain the deviation value B at the time of the fault. i (j),

[0134] If the deviation value B i If (j) is greater than the preset deviation threshold, then the i-th sensor is faulty.

[0135] Step 8: Use the reconstructed value as the output parameter of the faulty sensor to achieve soft repair of the faulty sensor.

[0136] Verification of the effectiveness of the processing method provided in this embodiment:

[0137] (1) Detection of abnormal states of normal signals

[0138] Assuming the drone swarm is functioning normally and all sensors are operating without faults, to verify the effectiveness of the method described in this paper, the ADS output results of three drones under steady-state signals and different phase transition conditions are used. The response values ​​of each sensor output under different phase transitions are collected to construct a test sample as follows: Figure 3 As shown, before a phase transition occurs, its output value can be considered as a steady-state signal. Then, the fault detection model based on principal component analysis is used to analyze the above signal to determine whether it can effectively distinguish between normal and fault signals under non-fault conditions.

[0139] Based on the fault detection principle of principal component analysis, the cumulative variance percentage of the first principal component is 58.6%, which is less than the set threshold of 95%. The cumulative variance percentage of the first two principal components is 96.97%, which exceeds the set threshold. Therefore, the number of principal components is selected as 2. The reason for the certain correlation between the sensors is the presence of identical gas sensors in the system. Within the residual space, the threshold of the SPE statistic can be calculated when the confidence level of the Gaussian distribution is 95%. Figure 3 By analyzing the test samples shown, the SPE statistic can be calculated. Figure 4 As shown, principal component analysis can effectively and accurately identify both stationary and normal abrupt signals without false alarms.

[0140] (2) Abnormal state detection under fault conditions

[0141] When a sensor in the system malfunctions, in addition to indicating whether a system malfunction has occurred, it is also necessary to effectively isolate and restore the fault, i.e., to provide the location of the system fault and the best estimated value. Taking the power failure of the carbon monoxide sensor CO-1 in ADS as an example, a simulation of fault detection, isolation, and restoration is performed to verify the feasibility of the method used in this paper.

[0142] like Figure 5 As shown, taking a power outage fault as an example, sensor CO-1 experiences a power outage fault at 100 seconds, and its measured value rapidly drops from 2.5V to 0V. At this time, as... Figure 6 As shown, its SPE statistic rises rapidly and exceeds the set threshold SPE. α This indicates that one or more faults exist in the sensor system at this time. To further confirm the details of the faults, moving window principal component analysis is used for further fault analysis. Figure 5 The signal in the image is further decomposed, and a moving window method is applied for feature extraction. The results are as follows: Figure 7 As shown, the fault detection result obtained using this method still indicates that a fault exists in the cluster.

[0143] like Figure 8 As shown, to determine the fault location of the sensor, the PCA reconstruction method was applied to reconstruct the data and obtain the system deviation value at the time of the fault. At this time, the deviation value of sensor CO-1 was much larger than the expected value, proving that sensor CO-1 had a serious fault, while the deviation values ​​of other sensors were all within the normal error allowable range, and therefore were all in normal working condition. The specific deviation values ​​in the deviation value vector at this time are shown in Table 1 below.

[0144] Table 1. Deviation values ​​of each sensor in the sensor system under CO-1 power failure.

[0145] Table 1Bias value of each sensor in sensor system under outage faultof CO-1

[0146]

[0147] Example 2:

[0148] This embodiment provides a sensor system fault handling system for unmanned aerial vehicles (UAVs), including:

[0149] The data acquisition module is used to acquire the current output parameters of the UAV's sensor system and generate training samples;

[0150] The first principal component analysis module is used to standardize the training samples and perform dimensionality reduction based on principal component analysis.

[0151] The initial fault detection module is used to determine whether there is a fault in the drone's sensor system based on the training samples processed by dimensionality reduction;

[0152] The data architecture module is used to construct test samples from the current output parameters using the moving window method if a fault exists.

[0153] The second principal component analysis module is used to standardize the test samples and perform dimensionality reduction based on principal component analysis.

[0154] The fault re-examination module is used to re-determine whether the drone's sensor system is currently faulty based on the test samples after dimensionality reduction processing;

[0155] The fault isolation module is used to reconstruct the test sample if a fault exists, and to find the faulty sensor based on the reconstructed values.

[0156] The soft repair module is used to use the reconstructed value as the output parameter of the faulty sensor, thereby enabling soft repair of the faulty sensor.

[0157] To address the problem of unmanned aerial vehicle (UAV) swarms struggling to effectively detect their own status in complex environments, under electromagnetic interference, or in sudden situations, this invention proposes a sensor system fault handling method and system for UAV swarms. This system enables three core functions of the UAV swarm: 1. Online fault detection of the output values ​​of each UAV sensor during system operation; 2. Identification of which UAV(s) and which sensors(s) have failed when a fault is detected; 3. Reconstruction of the faulty sensor's data and provision of its normal output value, achieving soft fault recovery.

[0158] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can 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, etc.) containing computer-usable program code.

[0159] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0160] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0161] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0162] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for handling sensor system faults in a drone swarm, characterized in that, include: Obtain the current output parameters of the sensor system for each drone and generate training samples; The training samples are standardized and then dimensionality is reduced based on principal component analysis. Determine whether the drone's sensor system is currently faulty based on the training samples processed by dimensionality reduction; If a fault exists, the current output parameters will be used to construct a test sample using the moving window method; The test samples were standardized and then dimensionality was reduced based on principal component analysis. Based on the dimensionality reduction of the test samples, we can further determine whether the drone's sensor system is currently faulty; If a fault is found, the test sample is reconstructed, and the faulty sensor is located based on the reconstructed values. The reconstructed value is used as the output parameter of the faulty sensor, thereby enabling soft repair of the faulty sensor; The method of determining whether the UAV's sensor system is currently faulty based on the training samples processed by dimensionality reduction includes: The dimensionality-reduced training samples are transformed into the projection matrix C of the principal space and the projection matrix of the residual space. In the process, the principal space and residual space are obtained, denoted as […]. and These are the standardized training samples; in, For the dimensionality-reduced training samples, C = PP T , P is the principal element load matrix. Let I be the residual loading matrix. n×n It is an n-dimensional identity matrix; The SPE statistic, calculated based on the squared prediction error, shows the variation of the dimensionality-reduced training samples within the residual space. Determine if the SPE statistic matches the preset SPE threshold. α size, If SPE≥SPE α This proves that the sensor system is faulty; If SPE <SPE α This proves that the sensor system is not faulty; The sensor that reconstructs the test sample and locates the fault based on the reconstructed values ​​includes: The optimal reconstruction value of the sensor is obtained by using multiple iterations: Reconstructing the test samples based on the principal component space: The iterative expression is: Where, x n (j) represents the output parameter of the nth sensor at time j when the sensor is faulty. C j c is the j-th column vector of the projection matrix of the principal element space. jj Let x be the projection matrix parameter of the j-th sensor at time j. jj This represents the output parameter indicating that the j-th sensor malfunctions at time j. Let be the reconstructed value of the output parameter of the i-th sensor at time j; j = 1, 2, ..., n. Let x be the output parameter of the i-th sensor. i The vector consisting of the first j-1 variables, Let x be the output parameter of the i-th sensor. i A vector consisting of the last nj variables; when Approaching Upon convergence, the reconstruction formula is obtained: Calculate the difference between the output parameters of each sensor in the sensor system and the reconstructed values ​​of the output parameters, and obtain the deviation value B at the time of the fault. i (j), If the deviation value B i If (j) is greater than the preset deviation threshold, then the i-th sensor is faulty.

2. The method for handling sensor system faults in a drone swarm according to claim 1, characterized in that, The process of acquiring the current output parameters of each UAV's sensor system and generating training samples includes: Acquire the output parameters of each sensor when the drone's sensor system is working normally; Training samples are generated using the number of output parameters of the sensors as the number of rows and the number of sensors as the number of columns.

3. The method for handling sensor system faults in a drone swarm according to claim 1, characterized in that, The standardization of training samples and the dimensionality reduction based on principal component analysis include: Standardize the training samples: Where X represents the training samples. For the standardized training samples, E(X) is the mean vector of the training samples X, denoted as E(X) = [μ1, μ2, ..., μ...]. n ], μ n Let D be the mean of the output parameters of the nth sensor. σ Let X be the variance matrix of the training samples, denoted as σ n Let be the standard deviation of the output parameters of the nth sensor, 1 m It is an m-dimensional column vector with all elements equal to 1, where m is the number of output parameters of the sensor; Principal component analysis was performed on the standardized training samples to obtain their correlation coefficient matrix R: Singular value analysis was performed on the correlation coefficient matrix R: R=UΛU T Where Λ is the characteristic matrix of R, denoted as Λ=diag(λ i (i = 1, 2, ..., n), λ i U is the eigenvalue; U is the eigenvector of R, denoted as U = [u1, u2, ..., u] n ], u n It is a column vector; Take the first k dimensions of the eigenvector U as linearly independent vectors, where k < n, denoted as P = [u1, u2, …, u k , to form the principal component loading matrix; take the (k + 1)-th to n-th dimensions of the eigenvector U as linearly independent vectors, denoted as to form the residual loading matrix; The value of the principal component number k is determined based on the cumulative variance percentage principle; Dimensionality reduction is performed on the standardized training samples based on the number of principal components k.

4. The method for handling sensor system faults in a drone swarm according to claim 3, characterized in that, The method for determining the number of principal components k based on the cumulative variance percentage principle includes: Calculate the cumulative contribution rate of the principal component loading matrix: Where, λ j The eigenvalues ​​of the principal element loading matrix are given by R, and trace(R) is the trace of the principal element loading matrix. The number of principal components k is calculated based on the preset CPV(k) value.

5. The method for handling sensor system faults in a drone swarm according to claim 1, characterized in that, The preset threshold SPE α for: Where α is the validation level, C α For the corresponding standard normal distribution, For parameter θ i The j-th feature is given by k, the number of principal components is given by n, and the number of sensors is given by n.

6. The method for handling sensor system faults in a drone swarm according to claim 1, characterized in that, The step of constructing test samples using the moving window method based on the current output parameters includes: The current output parameter of the sensor system is added to the adjacent s-1 output parameters to form the test data sample x. * ; x * =x ij +x i(j+1) +…+x i(j+s-1) +Δb ij +Δb i(j+1) +…+Δb i(j+s-1) Where s represents the window length, x ij Δb represents the output parameter where the i-th sensor fails at time j. ij For the output parameter x ij Fault deviation; Since the average value of the adjacent s-1 output parameters is close to 0 after standardization, the test data sample x * =x ij +Δb ij +Δb i(j+1) +…+Δb i(j+s-1) As a test sample.

7. A sensor system fault handling system for a drone swarm, characterized in that, The system is used to implement the steps of the method as described in any one of claims 1-6, the system comprising: The data acquisition module is used to acquire the current output parameters of the UAV's sensor system and generate training samples; The first principal component analysis module is used to standardize the training samples and perform dimensionality reduction based on principal component analysis. The initial fault detection module is used to determine whether there is a fault in the drone's sensor system based on the training samples processed by dimensionality reduction; The data architecture module is used to construct test samples from the current output parameters using the moving window method if a fault exists. The second principal component analysis module is used to standardize the test samples and perform dimensionality reduction based on principal component analysis. The fault re-examination module is used to re-determine whether the drone's sensor system is currently faulty based on the test samples after dimensionality reduction processing; The fault isolation module is used to reconstruct the test sample if a fault exists, and to find the faulty sensor based on the reconstructed values. The soft repair module is used to use the reconstructed value as the output parameter of the faulty sensor, thereby enabling soft repair of the faulty sensor.

Images