CN115077594B - Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism - Google Patents

Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism Download PDF

Info

Publication number
CN115077594B
CN115077594B CN202210949550.2A CN202210949550A CN115077594B CN 115077594 B CN115077594 B CN 115077594B CN 202210949550 A CN202210949550 A CN 202210949550A CN 115077594 B CN115077594 B CN 115077594B
Authority
CN
China
Prior art keywords
unmanned aerial
aerial vehicle
neighbor
drone
state
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210949550.2A
Other languages
Chinese (zh)
Other versions
CN115077594A (en
Inventor
程月华
胡恒嵩
姜斌
余自权
董凌霄
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202210949550.2A priority Critical patent/CN115077594B/en
Publication of CN115077594A publication Critical patent/CN115077594A/en
Application granted granted Critical
Publication of CN115077594B publication Critical patent/CN115077594B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C25/00—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Manufacturing & Machinery (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Automation & Control Theory (AREA)
  • Traffic Control Systems (AREA)

Abstract

The utility model discloses a bee colony unmanned aerial vehicle fault detection method based on LSTM and neighbour trust mechanism, relate to the fault detection technology field, this method constructs the prediction model based on LSTM network, neighbour unmanned aerial vehicle can predict the sensor output value of next moment in real time and obtain the state predicted value according to sensor historical data, combine the state observed value can obtain neighbour's testing result, and add neighbour trust mechanism and synthesize neighbour's testing result and confidence weight and obtain final fault detection result, this method need not to carry out complicated bee colony unmanned aerial vehicle dynamics model linearization, can be effective, whether the detection bee colony unmanned aerial vehicle breaks down in the sensor fast, effectively avoid the harm that brings to whole cluster by the sensor trouble, be applicable to very much that bee colony unmanned aerial vehicle dynamics model linearization is difficult and have the scene of many parameter couplings, provide the theoretical foundation for effectively avoiding the harm that the sensor trouble produced to bee colony unmanned aerial vehicle.

Description

Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism
Technical Field
The application relates to the technical field of fault detection, in particular to a swarm unmanned aerial vehicle fault detection method based on an LSTM and a neighbor trust mechanism.
Background
Unmanned Aerial Vehicle (Unmanned Aerial Vehicle) refers to an Aerial intelligent agent which can realize flight control in a remote control or autonomous control mode without artificial driving control or Unmanned Aerial Vehicle, and is mainly divided into a fixed-wing Unmanned Aerial Vehicle and a rotor Unmanned Aerial Vehicle. The fixed wing unmanned aerial vehicle has long endurance and good wind resistance. Rotor unmanned aerial vehicle is with low costs, flexible, can hover at the fixed point of motion. The unmanned aerial vehicle is widely applied to various industries by virtue of the advantages, particularly in military mission application scenes, and the fixed-wing unmanned aerial vehicle has excellent performance in missions such as battlefield reconnaissance, communication relay and information countermeasure by virtue of the long-distance flight characteristic of the fixed-wing unmanned aerial vehicle.
Due to the limited function of a single unmanned aerial vehicle, more complex tasks are difficult to complete, and a plurality of biological populations in nature can present certain group behaviors through mutual communication and cooperation among individuals. People are inspired, and a single unmanned aerial vehicle is connected by using a wireless communication network to form an intelligent unmanned aerial vehicle cluster. Mutual cooperation is realized through information interaction of the unmanned aerial vehicles in the cluster, and the success rate of completing complex tasks is greatly improved. The swarm unmanned aerial vehicle is an intelligent unmanned aerial vehicle cluster based on a swarm control strategy. The swarm unmanned aerial vehicle has the advantages of low cost, easiness in deployment, cluster diversity and the like, and is applied to the military and civil fields. The land-air cooperative fixed wing swarm unmanned aerial vehicle system developed by the middle-electric department switches formation in the air flexibly through the unmanned aerial vehicle, and achieves the maximum striking effect through multi-round striking.
Unmanned aerial vehicle sensor is comparatively accurate mechanical device usually, and operational environment is relatively poor, in order to improve structure integration level, reduce cost moreover. At present, many unmanned aerial vehicles adopt non-redundancy or low-redundancy design, sensors become weak links of a flight control system, and the fault occurrence rate is high. Compare single unmanned aerial vehicle and take place the sensor trouble, bee colony unmanned aerial vehicle's trouble individual can propagate the influence of trouble to neighbor unmanned aerial vehicle through communication network, influences the flight of whole cluster. Therefore, accurate and rapid realization of the fault detection of the sensor of the swarm unmanned aerial vehicle is the key for ensuring the flight reliability of the swarm unmanned aerial vehicle.
At present, a detection method for the fault of the sensor of the unmanned aerial vehicle mainly comprises a detection method based on a model and a detection method based on signal processing:
(1) The detection method based on the model commonly uses a distributed observer and a sliding-mode observer. However, the model-based detection method relies on an accurate mathematical model and is not suitable for complex systems. And because bee colony unmanned aerial vehicle system does not have fixed formation, the unmanned aerial vehicle only carries out the information interaction with the neighbour's unmanned aerial vehicle in the communication range, can't establish the exact mathematical model of bee colony unmanned aerial vehicle system, consequently is difficult to adopt the mode of model to carry out failure analysis to bee colony unmanned aerial vehicle.
(2) The detection method based on signal processing is usually used for mining fault characteristics in data by methods such as wavelet transformation, modal decomposition and the like. However, the signal processing-based method only analyzes single variable time domain or frequency domain information, neglects the coupling relationship among multiple variables, has low accuracy of fault analysis, and is difficult to satisfy the multi-parameter coupling scene of the swarm unmanned aerial vehicle.
Therefore, the existing detection method based on the model and the detection method based on the signal processing are difficult to accurately and effectively realize the fault detection of the unmanned aerial vehicle sensor.
Disclosure of Invention
The applicant provides a swarm unmanned aerial vehicle fault detection method based on an LSTM and a neighbor trust mechanism aiming at the problems and technical requirements, and the technical scheme of the method is as follows:
a method for detecting faults of swarm unmanned aerial vehicles based on LSTM and neighbor trust mechanism comprises the following steps for any unmanned aerial vehicle i in the swarm unmanned aerial vehicles:
each neighbor unmanned aerial vehicle j of the unmanned aerial vehicle i respectively acquires a state observation value of the unmanned aerial vehicle i at the detection time t;
each neighbor unmanned aerial vehicle j obtains a state prediction value of the neighbor unmanned aerial vehicle j aiming at the unmanned aerial vehicle i at the detection time t by using a prediction model according to historical state data of the unmanned aerial vehicle i within a preset time before the detection time t, and the prediction model is obtained based on LSTM network training;
each neighbor unmanned aerial vehicle j generates a neighbor detection result of the unmanned aerial vehicle i at the detection time t according to the acquired state observation value and the acquired state prediction value, and the neighbor detection result is used for indicating the neighbor unmanned aerial vehicle j to judge whether the unmanned aerial vehicle i is in a fault state or a normal state;
and performing weighted calculation on the neighbor detection result of each neighbor unmanned aerial vehicle j at the detection time t according to the trust weight of each neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i to obtain the final fault detection result of the unmanned aerial vehicle i at the detection time t, wherein the final fault detection result is used for indicating that the unmanned aerial vehicle i is in a fault state or in a normal state.
The further technical scheme is that the trust weight of each neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i is related to the network centrality of the neighbor unmanned aerial vehicle j and/or the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i;
the higher the network centrality is, the higher the trust level weight of the neighbor unmanned aerial vehicle j is when generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a fault state, and the lower the trust level weight is when generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a normal state;
the higher the reliability of communication with the drone i, the higher the trust level weight of the neighbor drone j.
The further technical scheme is that the method also comprises the following steps:
and each neighbor unmanned aerial vehicle j updates the trust degree weight of the neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i at each detection moment according to the network centrality of the neighbor unmanned aerial vehicle j and/or the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i.
The further technical scheme is that the confidence weight of each neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i at the detection time t
Figure BDA0003788867660000031
Comprises the following steps:
Figure BDA0003788867660000032
wherein alpha is 1 、α 2 Are all weighted, α 1 ∈[0,1]、α 2 ∈[0,1]、α 1 +α 2 ＝1；
Figure BDA0003788867660000033
Represents the neighbor detection result of the neighboring drone j on drone i when ≥ h>
Figure BDA0003788867660000034
The time indicates the neighbor unmanned plane j to judge that the unmanned plane i is in a normal state, and when ^ is in a normal state>
Figure BDA0003788867660000035
Indicating a neighbor unmanned aerial vehicle j to judge that the unmanned aerial vehicle i is in a fault state; ME' j (t) is a parameter obtained by normalizing the network centrality of the neighboring drone j at the detection time t, and ME 'is determined as the network centrality of the neighboring drone j is higher' j The more (t)Large; />
Figure BDA0003788867660000036
Represents a normalized parameter determined based on the communication reliability between the neighboring drone j and the drone i at the detection time t.
The further technical scheme is that the network centrality ME of the neighbor unmanned aerial vehicle j j Comprises the following steps:
Figure BDA0003788867660000037
wherein, DC j Is the standardization centrality of the neighbor drone j and
Figure BDA0003788867660000038
the swarm unmanned aerial vehicle comprises N unmanned aerial vehicles, d j Is the degree of the neighbor drone j and represents the total number of direct contacts between the neighbor drone j and other N-1 drones in the swarm drone; the neighbor unmanned aerial vehicle j has M neighbor unmanned aerial vehicles in total, and the standardization centrality of any mth neighbor unmanned aerial vehicle of the neighbor unmanned aerial vehicle j is DC m 。
The further technical proposal is that the communication reliability between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i is expressed based on the communication energy consumption between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i,
Figure BDA0003788867660000039
the communication energy consumption between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i is normalized to obtain a parameter; the lower the communication energy consumption, the higher the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i.
The further technical scheme is that the communication energy consumption between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i
Figure BDA00037888676600000310
Comprises the following steps:
Figure BDA00037888676600000311
wherein the distance between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i is d, the transmitted data volume is kbit,
Figure BDA00037888676600000312
represents the energy consumption for transmitting kbit data in a communication link at a distance d, is->
Figure BDA00037888676600000313
Representing the energy consumption of the power amplifier in a multipath fading model; e elec Is the energy consumption of the unmanned aerial vehicle communication circuit sensing 1bit data, epsilon fs The energy consumption of the power amplifier in the free space fading model for processing kbit data is shown, and r is a wireless channel constant.
The prediction model comprises a plurality of stages of sequentially cascaded LSTM networks, and a Droupout layer is arranged between two adjacent cascaded LSTM networks.
The further technical scheme is that the method for obtaining the state prediction value of the unmanned aerial vehicle i by each neighbor unmanned aerial vehicle j comprises the following steps:
and the neighbor unmanned aerial vehicle j inputs historical state data of the unmanned aerial vehicle i within a preset time before the detection time t into the prediction model, the historical state data comprises a position time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction and a speed time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction, and the predicted position of the unmanned aerial vehicle i in the three-dimensional motion direction at the detection time t output by the prediction model is used as a state predicted value of the unmanned aerial vehicle i.
The further technical scheme is that the method for generating the neighbor detection result of the unmanned aerial vehicle i at the detection moment t by each neighbor unmanned aerial vehicle j comprises the following steps:
the neighbor unmanned aerial vehicle j calculates a residual error result of the state observation value and the state prediction value of the unmanned aerial vehicle i at the detection moment t, and if the residual error result exceeds a preset threshold value, a neighbor detection result indicating that the unmanned aerial vehicle i is in a fault state is generated; and if the residual error result does not exceed the preset threshold, generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a normal state.
The beneficial technical effect of this application is:
the utility model discloses a bee colony unmanned aerial vehicle fault detection method based on LSTM and neighbour trust mechanism, this method constructs prediction model based on LSTM network, can predict the sensor output value of next moment in real time and receive the state prediction value according to sensor historical data, combine the state observation value can obtain the neighbour detection result, and add neighbour trust mechanism and synthesize neighbour detection result and confidence weight and obtain final fault detection result, this method need not to carry out complicated bee colony unmanned aerial vehicle dynamics model linearization, can effectively, detect whether the bee colony unmanned aerial vehicle takes place the sensor trouble fast, effectively avoid the harm that brings to whole cluster by the sensor trouble, be applicable to very much that bee colony unmanned aerial vehicle dynamics model linearization is difficult and have the scene of many parameter coupling, provide the theoretical foundation for effectively avoiding the harm that the sensor trouble produced to bee colony unmanned aerial vehicle.
According to the neighbor trust mechanism, the network centrality and neighbor communication energy consumption of the unmanned aerial vehicle are used as beam indexes of the trust degree weight, the neighbor trust mechanism reduces the influence of misjudgment and missed judgment caused by sensor measurement disturbance and the like, and the reliability of the algorithm is improved.
Drawings
Fig. 1 is a flow chart of each neighbor drone j generating a neighbor detection result for drone i in one embodiment of the present application.
Fig. 2 is a flowchart of obtaining a final failure detection result of drone i from neighbor detection results of each neighbor drone j in an embodiment of the present application.
Fig. 3 is an initial position profile of all drones in one example.
Fig. 4 is a graph showing the distribution of the speeds of all the drones in the example shown in fig. 3 in the three-dimensional movement direction.
Fig. 5 is a value of the network centrality of all drones in the example shown in fig. 3.
FIG. 6 is a graph of loss values of the predictive model for the training set and the test set as a function of epoch for the example shown in FIG. 3.
Fig. 7 is a schematic graph of the predicted state value and the observed state value of the neighboring drone of the drone No. 8 in the example shown in fig. 3 for the position of the drone No. 8 in the x direction.
Fig. 8 is a graph showing changes in the residual results calculated from the state prediction values and the state observation values in fig. 7.
Fig. 9 is a graph showing the variation of the final fault detection result of the unmanned aerial vehicle No. 8.
Detailed Description
The following description of the embodiments of the present application will be made with reference to the accompanying drawings.
The application discloses a swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism, which comprises the following steps, please refer to FIG. 1 and FIG. 2:
step S1, each neighbor unmanned aerial vehicle j of the unmanned aerial vehicle i respectively acquires a state observation value of the unmanned aerial vehicle i at the detection time t. In the system that bee colony unmanned aerial vehicle constitutes, every unmanned aerial vehicle's neighbour unmanned aerial vehicle contains other unmanned aerial vehicle in its perception scope, and every unmanned aerial vehicle carries out communication interaction with its each neighbour unmanned aerial vehicle, and can be through the sensor perception neighbour unmanned aerial vehicle's that the machine carried state information.
In one embodiment, the state observation value of the drone i acquired by each neighboring drone j is the position of the drone i in the three-dimensional motion direction, which is actually measured by the sensor.
And S2, obtaining a state prediction value of the neighboring unmanned aerial vehicle j for the unmanned aerial vehicle i at the detection time t by using a prediction model according to historical state data of the unmanned aerial vehicle i within a preset time before the detection time t.
Wherein, the prediction model is obtained by pre-training based on the LSTM network. And the neighbor unmanned aerial vehicle j inputs historical state data of the unmanned aerial vehicle i within a preset time length before the detection time t into a pre-trained prediction model, the historical state data comprises a position time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction and a speed time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction, and the predicted position of the unmanned aerial vehicle i in the three-dimensional motion direction at the detection time t, which is output by the prediction model, is used as a state prediction value of the unmanned aerial vehicle i.
The unmanned aerial vehicle of bee colony to do not have fixed formation restraint and neighbour's quantity, unmanned aerial vehicle is loose, the distribution state of coupling in the space, and every unmanned aerial vehicle is interactive with other unmanned aerial vehicles that relative distance is less than perception distance. The overall consistency is realized through local information interaction between the swarm unmanned aerial vehicles, the speed is consistent, and the relative distance is kept unchanged. The unmanned aerial vehicle trouble that this application was considered is the sensor trouble, and the sensor trouble type of considering be the deviation trouble, and the reason that this type of trouble takes place is that the change of temperature produces bias current or voltage for a constant bias is superimposed in sensor output, and this trouble mathematical model is:
Figure BDA0003788867660000061
/>
wherein t is a time variable, t s D is a constant offset superimposed when the sensor has an offset failure, and is a constant relating to the degree of the offset failure. y is sensor Is the value output by the sensor, and y (t) is the actual sensing value of the sensor at the detection time t. Sensor faults make the drone unable to measure correct data, thereby producing a false control rate, making the drone deviate from group consistency. Swarm unmanned aerial vehicle controls based on bee-crowd control strategy, and the control law of every unmanned aerial vehicle is not only relevant to the state parameter of self, still relevant with neighbor unmanned aerial vehicle's state parameter, so in the system that swarm unmanned aerial vehicle constitutes, unmanned aerial vehicle's sensor trouble brings the influence not only can influence unmanned aerial vehicle itself, still can influence its neighbor unmanned aerial vehicle simultaneously, makes the trouble propagate in swarm unmanned aerial vehicle.
Because the swarm unmanned aerial vehicle system does not have a fixed formation, the individuals only carry out information interaction with the individuals in the communication range, an exact mathematical model of the swarm unmanned aerial vehicle system cannot be established, and the faults are difficult to analyze in a model mode. Therefore, the fault detection method based on knowledge without establishing a complex model is adopted, the fault detection is carried out by utilizing an artificial intelligence algorithm based on the detection method based on knowledge without establishing a complex model, the multivariate coupling relation can be mined, and the fault detection method is more suitable for the fault detection of the swarm unmanned aerial vehicle. Considering that the sensor fault detection of the swarm unmanned aerial vehicle is aimed at, and the sensor data has characteristics of correlation and non-point information isolated from time, a regression method is used for searching the relation of parameters in time. Therefore, the prediction model is constructed and trained by adopting the LSTM network (improved recurrent neural network), the algorithm can well process time sequence data, and the algorithm well solves the problems of gradient loss and gradient explosion of the traditional recurrent neural network due to the unique internal structure of the algorithm.
In the conventional univariate prediction, only the historical data and time of the univariate prediction can be referred to, and some complex prediction situations are often determined by multiple factors, so that the univariate prediction is simple. Meanwhile, for the prediction of multiple objects, a plurality of univariate prediction models are constructed, so that the prediction structure is bloated. According to the method and the device, a prediction model is built based on the LSTM network, so that the prediction model has the characteristic of multi-input and multi-output, and the scene needs of fault detection of the swarm unmanned aerial vehicle are met.
In one embodiment, the predictive model is built based on a single LSTM network. Or in another embodiment, the prediction model comprises a plurality of stages of LSTM networks which are cascaded in sequence to increase the nonlinear fitting capability of the prediction model, and a Droupout layer is arranged between two adjacent cascaded LSTM networks for limiting the complexity of the prediction model and preventing the model from being over-fitted.
And S3, each neighbor unmanned aerial vehicle j generates a neighbor detection result of the unmanned aerial vehicle i at the detection time t according to the acquired state observation value and the acquired state prediction value.
And the neighbor detection result is used for indicating the neighbor unmanned aerial vehicle j to judge whether the unmanned aerial vehicle i is in a fault state or in a normal state. This application is to
Figure BDA0003788867660000071
Represents the neighbor detection result of the neighboring drone j on drone i when ≥ h>
Figure BDA0003788867660000072
The time indicates the neighbor unmanned plane j to judge that the unmanned plane i is in a normal state, and when ^ is in a normal state>
Figure BDA0003788867660000073
And then indicating the neighbor unmanned aerial vehicle j to judge that the unmanned aerial vehicle i is in a fault state.
Specifically, the method for generating the neighbor detection result comprises the following steps: calculating a residual error result of the state observed value and the state predicted value of the unmanned aerial vehicle i at the detection moment t by the neighbor unmanned aerial vehicle j
Figure BDA0003788867660000074
And if the residual error result exceeds a preset threshold value, generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a fault state. If the residual error result does not exceed the preset threshold, generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a normal state, namely ^ greater than or equal to>
Figure BDA0003788867660000075
And S4, carrying out weighted calculation on neighbor detection results of the neighbor unmanned aerial vehicles j at the detection time t according to the trust weights of the neighbor unmanned aerial vehicles j relative to the unmanned aerial vehicle i to obtain final fault detection results of the unmanned aerial vehicle i at the detection time t, wherein the final fault detection results are used for indicating that the unmanned aerial vehicle i is in a fault state or a normal state. That is, neighbor detection results of all neighbor drones j of the drone i are summarized to finally determine whether the drone i fails.
But there is unreasonable place to see all neighbor drone's neighbor detection results at once. When a certain neighbor drone receives wrong state information of the drone i due to a long distance and a large transmission data packet, a wrong neighbor detection result may be obtained, and if the neighbor detection result is equally weighted with neighbor detection results of other normal neighbor drones, a final fault detection result may be wrong, so that fault detection accuracy or false alarm rate is affected. Meanwhile, important nodes exist in the swarm unmanned aerial vehicle, and the topological position of the important nodes causes that the nodes can cause collision or divergence of the whole cluster once the nodes break down, so that more protection needs to be given to the important nodes. Therefore, a set of trust mechanism model needs to be established, the trust weight of each neighbor drone j relative to the drone i is determined, and weighted calculation is performed according to the trust weight.
In this application, the trust weight of each neighboring drone j with respect to drone i is related to the network centrality of the neighboring drone j and/or the reliability of the communication between the neighboring drone j and drone i. The neighbor drone j with higher network centrality has higher confidence weight when generating a neighbor detection result indicating that the drone i is in a fault state, and has lower confidence weight when generating a neighbor detection result indicating that the drone i is in a normal state. The higher the reliability of communication with drone i, the greater the confidence weight of neighbor drone j.
Based on the trust degree mechanism model, the higher the network centrality of the neighboring unmanned aerial vehicle j is, the more important the neighboring unmanned aerial vehicle j is, when the neighboring unmanned aerial vehicle j is judged to be in a fault state, the higher the trust degree weight is distributed to the neighboring unmanned aerial vehicle j, so that the neighboring unmanned aerial vehicle j is more inclined to believe the neighboring detection result of the neighboring unmanned aerial vehicle j, and even if the neighboring detection result of the neighboring unmanned aerial vehicle j as an important node is wrong, the safety can be ensured. And when the neighboring unmanned aerial vehicle j judges that the unmanned aerial vehicle i is in a normal state, the lower the trust level weight is distributed to the neighboring unmanned aerial vehicle j. In addition, the lower the reliability of communication between the neighboring drone j and the drone i is, which means that the neighboring drone j is more likely to obtain an erroneous neighboring detection result due to a transmission error, the lower the confidence weight assigned to the neighboring drone j is.
In an embodiment, taking the example of obtaining the trust weight of the neighboring drone j relative to the drone i based on the network centrality and the communication reliability at the same time, based on the trust mechanism model, the following calculation formula is constructed to determine the trust weight of each neighboring drone j relative to the drone i at the detection time t
Figure BDA0003788867660000081
Comprises the following steps:
Figure BDA0003788867660000082
wherein alpha is 1 、α 2 Are all weighted, α 1 ∈[0,1]、α 2 ∈[0,1]、α 1 +α 2 ＝1。
Figure BDA0003788867660000083
Represents the neighbor detection result of the neighboring drone j on drone i when ≥ h>
Figure BDA0003788867660000084
The time indicates the neighbor unmanned plane j to judge that the unmanned plane i is in a normal state, and when ^ is in a normal state>
Figure BDA0003788867660000085
And then indicating the neighbor unmanned aerial vehicle j to judge that the unmanned aerial vehicle i is in a fault state. ME' j (t) is a parameter obtained by normalizing the network centrality of the neighboring drone j at the detection time t, and ME 'is determined as the network centrality of the neighboring drone j is higher' j The larger (t) is. />
Figure BDA0003788867660000086
Represents a normalized parameter determined based on the communication reliability between the neighboring drone j and the drone i at the detection time t. Therefore, the distribution principle of the trust degree weight to be realized by the trust degree mechanism model can be realized.
Those skilled in the art will understand that, in practical application, the trust weight may be obtained based on the network centrality of the neighboring drone j, and may be according to
Figure BDA0003788867660000087
The calculation formula of (a) obtains the trust weight ^ of the neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i at the detection time t>
Figure BDA0003788867660000088
Or may only be weighted based on the reliability of the communication between the neighbor drone j and drone i, then may be based on £ and £>
Figure BDA0003788867660000089
The calculation formula of (a) obtains the trust weight ^ of the neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i at the detection time t>
Figure BDA00037888676600000810
/>
In the above calculation formula, at any detection time, the network centrality ME of the neighboring drone j j Comprises the following steps:
Figure BDA00037888676600000811
wherein, DC j Is the standardization centrality of the neighbor drone j and
Figure BDA00037888676600000812
the swarm unmanned aerial vehicle comprises N unmanned aerial vehicles, d j Is the degree of the neighbor drone j and represents its total number of direct contacts with the other N-1 drones in the swarm drone. The neighbor unmanned aerial vehicle j has M neighbor unmanned aerial vehicles in total, and the standardization centrality of any mth neighbor unmanned aerial vehicle of the neighbor unmanned aerial vehicle j is DC m Specific calculation formula and the above DC j Similarly, the present application is not described in detail.
In one embodiment, the reliability of communication between the neighboring drone j and the drone i is expressed based on the energy consumption of communication between the neighboring drone j and the drone i, the lower the energy consumption of communication, the higher the reliability of communication between the neighboring drone j and the drone i. Then
Figure BDA0003788867660000091
Is to the communication energy consumption between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i->
Figure BDA0003788867660000092
And carrying out normalization processing to obtain parameters.
At any detection moment, the energy consumption of communication between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i
Figure BDA0003788867660000093
Comprises the following steps:
Figure BDA0003788867660000094
wherein the distance between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i is d, the transmitted data volume is kbit,
Figure BDA0003788867660000095
represents the energy consumption for transmitting kbit data in a communication link at a distance d, is->
Figure BDA0003788867660000096
Representing the power consumption of the power amplifier in a multipath fading model. E elec Is the energy consumption of the unmanned aerial vehicle communication circuit sensing 1bit data, epsilon fs The energy consumption of the power amplifier in the free space fading model for processing kbit data is shown, and r is a wireless channel constant.
In determining a confidence weight for each neighbor drone j relative to drone i
Figure BDA0003788867660000097
Then, the neighbor detection result of each neighbor unmanned aerial vehicle j at the detection time t is judged>
Figure BDA0003788867660000098
The final fault detection result ^ at the detection time t of the unmanned aerial vehicle i is obtained through weighted calculation>
Figure BDA0003788867660000099
N i Is a set of all neighbor drones of drone i. Based on->
Figure BDA00037888676600000910
When G is defined as i And (t) generating a final fault detection result indicating that the unmanned aerial vehicle i is in a fault state when the t is more than or equal to 0, otherwise generating a final fault detection result indicating that the unmanned aerial vehicle i is in a normal state.
Please refer to the drawings2, set N of all neighboring drones of drone i i Totally including q neighbor unmanned aerial vehicles and marking as j in turn 1 、j 2 、…j q The neighbor detection results obtained by the method are sequentially
Figure BDA00037888676600000911
The trust weight of each neighbor drone relative to drone i is ≥ in order>
Figure BDA00037888676600000912
A schematic diagram of the resulting fault detection results is shown for illustration.
In one embodiment, the trust weight of each neighboring drone j relative to drone i may be a constant value, or more commonly, since the swarm drones are dynamically changing, each neighboring drone j updates the trust weight of the neighboring drone j relative to drone i at each detection time according to the network centrality of the neighboring drone j and/or the communication reliability between the neighboring drone j and drone i, and then performs weighted calculation on the neighbor detection result by using the updated trust weight at each detection time. The updating method is the same as the above calculation method, and is not described in detail in this application.
In the method, each unmanned aerial vehicle needs to use a loaded prediction model, and before the method is executed, the method further comprises the step of training the prediction model, and firstly, a network structure of the prediction model is built by using an LSTM network according to actual needs. A training set and a test set are then constructed to train the predictive model. The training prediction model needs to use a large amount of training sets, and the training sets are also required to meet the input structure of the LSTM network, and for the LSTM network with multiple inputs and multiple outputs, the input data and the output data are both in a matrix form. Thus the data sequence to obtain a non-malfunctioning drone is written as F 0 ＝{f 1 ,f 2 ,····,f n N is the length of the sequence, a proper step length L is selected to be less than or equal to n, a sliding window with the length equal to the step length L is constructed to slide on the data sequence, and a training set D = [ D ] =isobtained 1 ,D 2 ,····,D n-L+1 ] T The matrix form of the training set D is:
Figure BDA0003788867660000101
each data in the data sequence comprises the position and speed of the three-dimensional motion direction at the corresponding moment, and D = [ D ] for the training set 1 ,D 2 ,…,D n-L+1 ] T And carrying out normalization processing to obtain a training set D' so as to prevent the prediction effect from being influenced by different dimensions. And taking the training set D' after the normalization processing as the input of the prediction model, and taking the position of the three-dimensional motion direction at each moment as the output corresponding to the data sequence before the moment, thereby obtaining the prediction model through training. After model training, the indexes such as accuracy rate, recall rate and the like of the prediction model are calculated by using a test set, the construction mode of the test set is similar to that of the training set, and the step length P of the training set<n-L+1。
Based on the method provided by the present application, in one example, the speed and position of 24 drones are randomly initialized, and the initial position distribution map of each drone is shown in fig. 3. Set for No. 8 and No. 15 unmanned aerial vehicle to take place sensor deviation trouble, the deviation degree is the 30% of sensor measurement. The time of occurrence of the failure is set to t s =150s, the total simulation duration is set to 300s, and the changes in the speeds Vx, vy, and Vz of all the drone in the swarm drone in the three-dimensional movement directions (x direction, y direction, and z direction) are shown in fig. 4 as (a), (b), (c) in this order. The network centrality of each unmanned aerial vehicle in the swarm unmanned aerial vehicle at the initial moment is shown in fig. 5, the vertical axis sequentially represents unmanned aerial vehicles with the numbers of 1-24 from bottom to top, all the numbers are not shown in the figure, and the horizontal axis represents the value of the network centrality.
Setting the step length to be L =5, and setting a sliding window to obtain a training set and a testing set to obtain a prediction model. The loss values of the predictive model for the training set and the test set as a function of epoch are shown in fig. 6. As can be seen from the loss value, the prediction model can obtain better prediction value.
In this example, a neighbor drone of drone No. 8, based on the method provided by the present applicationA graph diagram of the state observed value and the state predicted value of the drone pair No. 8 is shown in fig. 7, where the x-direction positions are compared. As shown in fig. 8, it can be seen that the difference between the state observed value and the state predicted value begins to increase at the time of the failure occurrence, resulting in a sudden change in the residual result, so that the method of the present application can obtain a neighbor detection result. The variation curve of the final fault detection result of the unmanned aerial vehicle # 8 calculated in the above way is shown in fig. 9, and it can be seen that when the time is 152s, G is i (t) =0 can detect that No. 8 unmanned aerial vehicles are in the fault state, and the check-out time is shorter, and therefore the visible detection timeliness is higher.
Further in other examples, set up 6 different fault injection positions of group to bee colony unmanned aerial vehicle, inject the trouble into these three kinds of scenes of an unmanned aerial vehicle, two unmanned aerial vehicles and three unmanned aerial vehicles respectively, contain the condition of injecting the trouble into many adjacent and many non-adjacent unmanned aerial vehicles when injecting the trouble into many unmanned aerial vehicles. Each input mode is set with 40 tests under different initial conditions, and the results of the accuracy, the recall rate and the average detection time of the method are obtained through statistics as follows:
Figure BDA0003788867660000111
through statistics of 40 samples in each group, the method can achieve the accuracy rate of more than 95% on average, the recall rate of more than 97% on average, and the average detection time of not more than 3s, so that the method can well solve the problem of sensor fault detection of the swarm unmanned aerial vehicle.
What has been described above is only a preferred embodiment of the present application, and the present application is not limited to the above examples. It is to be understood that other modifications and variations directly derived or suggested to those skilled in the art without departing from the spirit and concepts of the present application are to be considered as included within the scope of the present application.

Claims (5)

1. A method for swarm drone fault detection based on LSTM and neighbor trust mechanism, the method comprising for any drone i in a swarm drone:
each neighbor unmanned aerial vehicle j of the unmanned aerial vehicle i respectively acquires a state observation value of the unmanned aerial vehicle i at the detection time t;
each neighbor unmanned aerial vehicle j obtains a state prediction value of the neighbor unmanned aerial vehicle j at the detection time t for the unmanned aerial vehicle i by using a prediction model according to historical state data of the unmanned aerial vehicle i within a preset time before the detection time t, wherein the prediction model is obtained based on LSTM network training;
each neighbor unmanned aerial vehicle j generates a neighbor detection result of the unmanned aerial vehicle i at the detection time t according to the acquired state observation value and the acquired state prediction value, wherein the neighbor detection result is used for indicating the neighbor unmanned aerial vehicle j to judge whether the unmanned aerial vehicle i is in a fault state or a normal state;
weighting and calculating the neighbor detection result of each neighbor unmanned aerial vehicle j at the detection time t according to the trust weight of each neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i to obtain the final fault detection result of the unmanned aerial vehicle i at the detection time t, wherein the final fault detection result is used for indicating that the unmanned aerial vehicle i is in a fault state or in a normal state;
the trust weight of each neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i is related to the network centrality of the neighbor unmanned aerial vehicle j and/or the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i; the higher the network centrality is, the higher the trust level weight of the neighbor drone j is when generating a neighbor detection result indicating that the drone i is in a fault state, and the lower the trust level weight is when generating a neighbor detection result indicating that the drone i is in a normal state; the higher the reliability of communication with the unmanned aerial vehicle i, the higher the weight of the trust degree of the neighbor unmanned aerial vehicle j;
confidence weighting of each neighbor drone j relative to drone i at detection time t
Figure FDA0004109770470000011
Comprises the following steps:
Figure FDA0004109770470000012
wherein alpha is 1 、α 2 Are all weighted, α 1 ∈[0,1]、α 2 ∈[0,1]、α 1 +α 2 ＝1；
Figure FDA0004109770470000013
Represents the neighbor detection result of the neighboring drone j on drone i when ≥ h>
Figure FDA0004109770470000014
Indicating the neighbor drone j to judge that the drone i is in a normal state when ^ is greater than ^ and ^ is greater than ^>
Figure FDA0004109770470000015
The neighbor unmanned aerial vehicle j is indicated to judge that the unmanned aerial vehicle i is in a fault state; ME' j (t) is a parameter obtained by normalizing the network centrality of the neighboring unmanned aerial vehicle j at the detection time t, and the higher the network centrality of the neighboring unmanned aerial vehicle j is, the higher the ME' j The larger (t); />
Figure FDA0004109770470000016
A normalization parameter representing a determination based on the communication reliability between the neighboring drone j and the drone i at the detection time t;
network centrality ME of neighbor UAV j j Comprises the following steps:
Figure FDA0004109770470000017
wherein, DC j Is the standardization centrality of the neighbor drone j and
Figure FDA0004109770470000021
the swarm unmanned aerial vehicle comprises N unmanned aerial vehicles, d j The degree of the neighbor unmanned plane j represents the direct contact total number of the neighbor unmanned plane j and other N-1 unmanned planes in the swarm unmanned plane; the neighbor drone j has M neighbor drones in total,any mth neighbor drone of neighbor drone j has a centrality of standardization DC m ；
The reliability of communication between the neighbor drone j and the drone i is expressed based on the energy consumption of communication between the neighbor drone j and the drone i,
Figure FDA0004109770470000022
the parameter is obtained by carrying out normalization processing on communication energy consumption between a neighbor unmanned aerial vehicle j and an unmanned aerial vehicle i; the lower the communication energy consumption is, the higher the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i is; communication energy consumption between neighbor unmanned aerial vehicle j and unmanned aerial vehicle i->
Figure FDA0004109770470000023
Comprises the following steps:
Figure FDA0004109770470000024
wherein the distance between the neighboring unmanned aerial vehicle j and the unmanned aerial vehicle i is d, the transmitted data volume is kbit,
Figure FDA0004109770470000025
represents the energy consumption for transmitting kbit data in a communication link at a distance d, is->
Figure FDA0004109770470000026
Representing the energy consumption of the power amplifier in a multipath fading model; e elec Is the energy consumption of the unmanned aerial vehicle communication circuit sensing 1bit data, epsilon fs The energy consumption of the power amplifier in the free space fading model for processing kbit data is shown, and r is a wireless channel constant.
2. The method of claim 1, further comprising:
and each neighbor unmanned aerial vehicle j updates the trust degree weight of the neighbor unmanned aerial vehicle j relative to the unmanned aerial vehicle i at each detection moment according to the network centrality of the neighbor unmanned aerial vehicle j and/or the communication reliability between the neighbor unmanned aerial vehicle j and the unmanned aerial vehicle i.
3. The method according to claim 1, wherein the prediction model comprises several cascaded LSTM networks, and a Droupout layer is disposed between two adjacent cascaded LSTM networks.
4. The method of claim 1, wherein the method for each neighbor drone j to derive the predicted value of the state of drone i comprises:
and the neighbor unmanned aerial vehicle j inputs historical state data of the unmanned aerial vehicle i within a preset time before the detection time t into the prediction model, the historical state data comprises a position time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction and a speed time sequence of the unmanned aerial vehicle i in the three-dimensional motion direction, and the predicted position of the unmanned aerial vehicle i in the three-dimensional motion direction at the detection time t output by the prediction model is used as a state predicted value of the unmanned aerial vehicle i.
5. The method of claim 1, wherein the method for each neighbor drone j to generate neighbor detection results for drone i at detection time t comprises:
the neighbor unmanned aerial vehicle j calculates a residual error result of the state observation value and the state prediction value of the unmanned aerial vehicle i at the detection moment t, and if the residual error result exceeds a preset threshold value, a neighbor detection result indicating that the unmanned aerial vehicle i is in a fault state is generated; and if the residual error result does not exceed the preset threshold value, generating a neighbor detection result indicating that the unmanned aerial vehicle i is in a normal state.
CN202210949550.2A 2022-08-09 2022-08-09 Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism Active CN115077594B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210949550.2A CN115077594B (en) 2022-08-09 2022-08-09 Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210949550.2A CN115077594B (en) 2022-08-09 2022-08-09 Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism

Publications (2)

Publication Number Publication Date
CN115077594A CN115077594A (en) 2022-09-20
CN115077594B true CN115077594B (en) 2023-04-18

Family

ID=83244192

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210949550.2A Active CN115077594B (en) 2022-08-09 2022-08-09 Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism

Country Status (1)

Country Link
CN (1) CN115077594B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117520950B (en) * 2024-01-04 2024-03-19 贵州大学 Multi-target UAV fault diagnosis method based on attention knowledge sharing network

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104867357B (en) * 2015-01-21 2017-03-01 中南大学 Multiple no-manned plane scheduling and mission planning method towards Emergency Response to Earthquake
CN105763396B (en) * 2016-04-12 2018-02-06 北京理工大学 Distributed multi agent real-time fault detection method based on neighbours' correlation behavior
CN105915388B (en) * 2016-05-31 2019-06-04 广东电网有限责任公司电力调度控制中心 A kind of Transducer-fault Detecting Method and system based on distributed network
US11314242B2 (en) * 2019-01-28 2022-04-26 Exxonmobil Research And Engineering Company Methods and systems for fault detection and identification
CN112929205B (en) * 2021-01-22 2022-05-10 南京航空航天大学 Swarm unmanned plane fault propagation method based on cellular automaton
CN113051723B (en) * 2021-03-12 2022-02-18 南京航空航天大学 Swarm unmanned aerial vehicle fault propagation analysis method based on time sequence network

Also Published As

Publication number Publication date
CN115077594A (en) 2022-09-20

Similar Documents

Publication Publication Date Title
CN111044052B (en) Unmanned aerial vehicle self-adaptive navigation system and method based on intelligent sensing
CN109067773A (en) A kind of vehicle-mounted CAN network inbreak detection method neural network based and system
CN115077594B (en) Swarm unmanned aerial vehicle fault detection method based on LSTM and neighbor trust mechanism
He et al. State-aware rate adaptation for UAVs by incorporating on-board sensors
US20210326677A1 (en) Determination device, determination program, determination method and method of generating neural network model
Tzes et al. Graph neural networks for multi-robot active information acquisition
Titouna et al. An online anomaly detection approach for unmanned aerial vehicles
Ma et al. Reliability analysis of power and communication network in drone monitoring system
CN114615143A (en) Elastic distributed safety monitoring method under multi-sensor-observation network
CN110825112B (en) Oil field dynamic invasion target tracking system and method based on multiple unmanned aerial vehicles
Signori et al. Channel-based trust model for security in underwater acoustic networks
He et al. Graph attention network-based fault detection for UAVs with multivariant time series flight data
Liu et al. Sensor fault diagnosis of GPS/INS tightly coupled navigation system based on state chi-square test and improved simplified fuzzy ARTMAP neural network
CN115167147B (en) Distributed fault diagnosis method for swarm unmanned aerial vehicle based on thrust residual error estimation
Hajshirmohamadi et al. Event-triggered fault detection for discrete-time linear multi-agent systems
CN115204241A (en) Deep learning tiny fault diagnosis method and system considering fault time positioning
Titouna et al. An efficient probabilistic model for anomaly prediction in aerial Ad Hoc networks
CN113068151A (en) Unmanned aerial vehicle swarm task reliability analysis method and system considering system failure
Wei et al. ROSIS: Resilience oriented security inspection system against false data injection attacks
Giorgiani do Nascimento et al. Quantifying uncertainty in Li-ion battery aging due to unknown usage with hybrid physics-informed neural networks
CN110366232A (en) Sensor transmissions energy control method for remote status estimation
Chi et al. Research on real-time fault detection of UAV flight control system sensor based on Riemannian manifold
CN117034020B (en) Unmanned aerial vehicle sensor zero sample fault detection method based on CVAE-GAN model
CN115047892B (en) Underwater unmanned cluster formation key node identification method based on group intelligence
CN114237207B (en) Multi-agent distributed fault diagnosis method under influence of communication noise

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant