Unmanned aerial vehicle attitude control anomaly detection method and system based on data driving

Abstract

The invention discloses an unmanned aerial vehicle attitude control anomaly detection method and system based on data driving, and belongs to the technical field of unmanned aerial vehicle detection.The method comprises the steps that firstly, a complementary filtering model and an LSTM model are trained based on sensor data and flight attitude data of a four-rotor unmanned aerial vehicle in a historical time period; the trained complementary filtering model and the trained LSTM model output first attitude data and second attitude data; determining an attitude data error based on the attitude data of the previous moment output by the LSTM model and the actual attitude data of the unmanned aerial vehicle at the corresponding moment, and performing weighted fusion on the first attitude data and the second attitude data at the current moment according to the attitude data error to obtain unmanned aerial vehicle attitude prediction data; and finally, determining a confidence interval according to the residual error of the real unmanned aerial vehicle attitude data in the unmanned aerial vehicle attitude prediction data and the historical data, and detecting the attitude of the unmanned aerial vehicle according to the confidence interval. The method effectively improves the accuracy and reliability of unmanned aerial vehicle attitude control anomaly detection, and has important practical significance for improving the safety and reliability of unmanned aerial vehicle flight.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle detection, and in particular to a data-driven unmanned aerial vehicle attitude control anomaly detection method and system. Background Art

[0002] In recent years, the rapid development of UAV technology has led to its wide application in military and civilian fields, and it has played an increasingly important strategic value. However, with the increase in the complexity of the UAV system structure and the difficulty of mission execution, UAV failures have occurred frequently, leading to serious safety accidents and huge economic losses. For UAVs, flight attitude control is the core technology to ensure that the UAV is stable and can be accurately controlled. As an indirect manifestation of the operating status of the UAV, flight data is an important basis for evaluating flight performance and monitoring the health of the system. Therefore, it is of great practical significance to carry out data-driven UAV attitude control anomaly detection, provide accurate attitude in real time, keep the quadcopter UAV in a stable flight attitude, and avoid unnecessary casualties and losses, which is of great practical significance for improving the safety and reliability of UAV flight. At present, common anomaly detection methods include prior knowledge-based, model-based, and data-driven anomaly detection methods.

[0003] The publication number is CN108960303A, which is a method for detecting anomalies in UAV flight data based on LSTM. It explores the establishment of a deep learning model based on data-driven control for UAV attitude control. This method reconstructs the phase space data of UAV telemetry data and establishes an LSTM model for prediction to check abnormal points and thus detect anomalies in UAV flight data; the publication number is CN117851946A, which is a method, system and medium for detecting anomalies in multivariate flight time series data. The historical time series data of UAV is input into the RNN-CNN model for training and model establishment, and then the real-time flight data is input into the prediction model for anomaly detection, which solves the current problem of difficulty in finding and locating anomalies in massive flight data and improves the ability to detect anomalies in aircraft flight data.

[0004] However, in the above invention technical solution, when processing high-noise data in the real world based on deep model training, although the deep learning model has powerful feature extraction and pattern recognition capabilities, it may show insufficient model matching, especially when there are a lot of unpredictable noise or outliers in the data. This lack of matching is usually manifested as a decrease in the generalization ability of the model, unstable prediction results, or increased sensitivity to abnormal data. Summary of the invention

[0005] In view of the impact of high noise on the abnormal results of deep learning model verification flight data in the prior art, and the problem of how to improve the cluster collaboration of a group of quadrotor drones, the present invention provides a data-driven drone attitude control anomaly detection method and system, which deploys the quadrotor drone flight mission on a drone cloud network platform based on a docker container, thereby realizing remote status monitoring and task scheduling of the drone.

[0006] The present invention is achieved through the following technical solutions: In a first aspect, the present application provides a data-driven method for detecting anomalies in attitude control of a UAV, comprising: Based on the sensor data and flight attitude data of the quad-rotor drone in the historical time period, a complementary filter model and an LSTM model are trained respectively, and the trained complementary filter model and LSTM model output the first attitude data and the second attitude data; The attitude data error is determined based on the attitude data of the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment. The first attitude data and the second attitude data at the current moment are weightedly fused according to the attitude data error to obtain the drone attitude prediction data. The confidence interval is determined according to the residual between the predicted UAV attitude data and the actual UAV attitude data in the historical data, and the attitude of the UAV is detected according to the confidence interval.

[0007] Preferably, the sensor data includes three-axis accelerometer data, three-axis gyroscope data and three-axis magnetometer data; The flight attitude data includes a pitch angle, a roll angle and a yaw angle.

[0008] Preferably, the complementary filtering model outputs the first posture data including: The sensor data in the navigation coordinate system is converted into the sensor data in the carrier coordinate system, the quantity product error of the sensor data generated in the process of converting the navigation coordinate system and the carrier coordinate system is eliminated, and the sensor data in the carrier coordinate system is converted into Euler angles to obtain the first posture data.

[0009] Preferably, the sensor data conversion method in the carrier coordinate system is as follows: The quaternion transformation matrix between the navigation coordinate system and the carrier coordinate system is determined, and the sensor data in the navigation coordinate system at each moment is converted into the sensor data in the carrier coordinate system using the quaternion transformation matrix.

[0010] Preferably, the method for determining the first posture data includes: A PI controller is used to eliminate the vector product error generated in the sensor data conversion process, complete the compensation correction of the sensor data, and then convert the obtained sensor data into the Euler angle form to obtain the first posture data.

[0011] Preferably, the training method of the LSTM model includes: The sensor data of the drone at the previous n moments and the corresponding posture data are input into the LSTM model as training data, and the LSTM model is sequenced in combination with the stochastic gradient descent method. The trained LSTM model outputs the second posture data.

[0012] Preferably, the weighted fusion method of the first posture data and the second posture data is:

[0013]

[0014]

[0015] in, It is the error between the UAV attitude data calculated by the complementary filtering prediction model at the previous moment and the actual attitude data; It is the error between the drone attitude data output by the LSTM prediction model at the last moment and the actual attitude data; is the weight coefficient of the complementary filtering model; is the weight coefficient of the LSTM model; Predict data for drone attitude; is the first posture data; Refers to the second posture data.

[0016] Preferably, the detecting the posture of the drone according to the confidence interval includes: The sensor data of the UAV are respectively input into the trained complementary filtering model and the trained LSTM model. The complementary filtering model outputs the first attitude data, and the LSTM model outputs the second attitude data. The first attitude data and the second attitude data are weightedly fused based on the error to obtain the UAV attitude prediction data at the current moment. The residual of the UAV attitude prediction data and the actual attitude data is calculated, and the residual is compared with the confidence interval. When the residual exceeds the confidence interval, the attitude data at the current time point is abnormal.

[0017] In the second aspect, the present application provides a data-driven drone attitude control anomaly detection system, which is characterized by comprising: A prediction module is used to train a complementary filter model and an LSTM model based on the sensor data and flight attitude data of the quadrotor drone in a historical time period, and the trained complementary filter model and LSTM model output the first attitude data and the second attitude data; The attitude fusion module is used to determine the attitude data error based on the attitude data at the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment, and to perform weighted fusion on the first attitude data and the second attitude data at the current moment according to the attitude data error to obtain the drone attitude prediction data; The diagnosis module is used to determine the confidence interval according to the residual between the UAV attitude prediction data and the real UAV attitude data in the historical data, and detect the attitude of the UAV according to the confidence interval.

[0018] In a third aspect, the present application provides a "cloud-network-end" drone cluster control system, characterized in that it includes a Kubernetes cluster and a ground station; The Kubernetes cluster includes multiple drones, each drone serves as a working node, and the ground station serves as a master node to control the working node. The above-mentioned data-driven drone attitude control anomaly detection method is deployed on each working node.

[0019] Compared with the prior art, the present invention has the following beneficial technical effects: The present application provides a data-driven method for detecting anomalies in the attitude control of unmanned aerial vehicles. By combining the complementary filtering model and the LSTM model, the technical solution can make full use of the advantages of the two models. The complementary filtering model usually has a good response and stability to real-time data, while the LSTM model is good at processing time series data and capturing long-term dependencies. This fusion helps to improve the matching degree and generalization ability of the model when facing high noise or outlier data, making the prediction results more stable and accurate. Secondly, the attitude data error is determined by the attitude data of the previous moment output by the LSTM model and the actual attitude data of the unmanned aerial vehicle at the corresponding moment, and the first attitude data and the second attitude data at the current moment are weighted and fused accordingly. This method of dynamically adjusting the weight can more accurately reflect the current true attitude of the unmanned aerial vehicle, thereby improving the accuracy of anomaly detection; in addition, the confidence interval is determined according to the residual of the unmanned aerial vehicle attitude prediction data and the real unmanned aerial vehicle attitude data in the historical data. This confidence interval provides a quantitative standard for the attitude detection of the unmanned aerial vehicle, making the abnormal judgment more objective and reliable. When the residual of the actual attitude data and the predicted data exceeds the confidence interval, it can be judged as abnormal, so that timely measures can be taken to avoid potential safety risks.

[0020] This application also proposes a data-driven UAV attitude control anomaly detection system, an electronic device and a computer storage medium, which have all the advantages of the above-mentioned data-driven UAV attitude control anomaly detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is the control architecture of the "cloud-network-terminal" drone cluster control system in the present invention; Figure 2 This is a flowchart of configuring the drone as a working node of the K8s cluster in the present invention; Figure 3 It is a flowchart of the first UAV attitude prediction of the complementary filtering model in the present invention; Figure 4 The figure is a flow chart of the method for detecting abnormality of attitude control of unmanned aerial vehicle in the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0025] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0026] A data-driven unmanned aerial vehicle attitude control anomaly detection method comprises the following steps: Step 1: Obtain the sensor data and flight attitude data of the quadrotor drone in the historical time period.

[0027] The sensor data includes three-axis accelerometer data, three-axis gyroscope data, and three-axis magnetometer data.

[0028] The flight attitude data includes Pitch (pitch angle), Roll (roll angle), Yaw (yaw angle) Step 2: training a complementary filtering model based on the sensor data and the flight attitude data, and the trained model predicts the first attitude data of the rotorcraft.

[0029] The sensor data in the navigation coordinate system obtained in step 1 is converted into sensor data in the carrier coordinate system, the quantity product error of the sensor data generated during the conversion process between the navigation coordinate system and the carrier coordinate system is eliminated, and the sensor data in the carrier coordinate system is converted into Euler angles to obtain the first posture data.

[0030] The sensor data conversion method in the carrier coordinate system is as follows: The quaternion transformation matrix between the navigation coordinate system and the carrier coordinate system is determined, and the sensor data in the navigation coordinate system at each moment is converted into the sensor data in the carrier coordinate system using the quaternion transformation matrix.

[0031] The method to eliminate the vector product error generated by the sensor data conversion is as follows: like Figure 3 As shown in FIG. 1 , the vector product error consists of two parts. One part is the vector product error obtained by converting the gravity acceleration from the navigation coordinate system to the gravity acceleration in the carrier coordinate system, and performing a vector product with the gravity acceleration in the carrier accelerometer. The other part is the vector product error obtained by converting the three-axis geomagnetic pole data in the carrier coordinate system at the current time point into the three-axis geomagnetometer data in the navigation coordinate system, and then converting it into standard geomagnetometer data. After converting the standard three-axis geomagnetometer data into the three-axis geomagnetometer data in the carrier coordinate system, performing a vector product with the original three-axis geomagnetometer data in the carrier coordinate system.

[0032] A PI controller is used to eliminate the vector product error generated in the sensor data conversion process, complete the compensation correction of the sensor data, and then convert the gyroscope data into the form of Euler angles for output to obtain the first attitude data of the quadrotor drone at the current time node.

[0033] Step 3: Train the LSTM model based on the sensor data and flight attitude data, and the trained model predicts the second attitude data of the rotorcraft.

[0034] During the training process of the LSTM model, the sensor data of the quadrotor drone at the previous n moments and the corresponding attitude data are input into the LSTM model as training data, and the LSTM model is sequenced in combination with the stochastic gradient descent method. The trained LSTM model outputs the second attitude data at the current moment.

[0035] The framework of the LSTM model is an input layer, a hidden layer, and an output layer. The input layer is divided into a training set and an experimental set. The hidden layer consists of 128 LSTM neural units, which are used to learn the relationship between the input features and the final model output results and perform state transfer. The output layer calculates the hidden state of the hidden layer input and the bias matrix and bias terms, and finally calculates the current posture angle through the activation function to output the final result.

[0036] Since the sensor data and attitude data of the drone are continuously collected time series data, LSTM can capture the long-term dependencies of the data through its internal memory unit, which is suitable for processing such continuously changing dynamic information. The model is trained using data and optimized using the stochastic gradient descent method.

[0037] Step 4: Determine the attitude data error based on the attitude data at the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment, and perform weighted fusion on the first attitude data and the second attitude data at the current moment according to the attitude data error to obtain the predicted attitude data of the drone at the current moment.

[0038] After the LSTM model outputs the predicted second posture data at the current moment and the complementary filtering model outputs the predicted first posture data at the current moment, the first posture data and the second posture data are weightedly fused based on the error at the previous moment to reduce the impact of noise in the UAV flight data on the UAV posture data.

[0039] After the error-based dynamic weighted fusion of the UAV attitude data output by the LSTM model and the UAV attitude data output by the complementary filter, the UAV attitude prediction data at the current moment is obtained. The calculation formula for dynamic weighted fusion is:

[0040]

[0041]

[0042] in, Refers to the error between the UAV attitude data calculated by the complementary filtering prediction model at the previous moment and the actual attitude data; Refers to the error between the drone attitude data calculated by the LSTM prediction model at the last moment and the actual attitude data; Refers to the weight coefficient of the complementary filtering prediction model under the weighted fusion formula at the current moment; Refers to the weight coefficient of the complementary filtering prediction model under the weighted fusion formula at the current moment; Refers to the current moment predicted drone attitude data value obtained through weighted fusion; Refers to the current moment predicted drone attitude data value obtained through the complementary filtering model; Refers to the current moment predicted drone attitude data value obtained through the LSTM model.

[0043] Since the LSTM model requires the UAV flight data of n moments as input, the UAV flight data of the first n moments cannot use the output results obtained by the LSTM model. Based on this, the UAV attitude data of the first n moments is obtained by complementary filtering.

[0044] Step 5: Determine a confidence interval based on the UAV attitude prediction data and the UAV attitude data in the historical data, and detect the UAV attitude based on the confidence interval.

[0045] Calculate the residuals between the predicted UAV attitude data and the actual UAV attitude data, and determine the confidence interval based on the residuals. The residuals contain random noise, which conforms to the normal distribution. The confidence interval (μ-3σ, μ+3σ) is obtained through the residuals. μ is the mean value of the residual data in the sample data, and σ is the standard deviation of the sample residuals in the sample set.

[0046] This application uses a drone attitude data anomaly detection algorithm that integrates complementary filtering and LSTM networks to extract and integrate data features such as drone sensors, equipment, and environment to support anomaly detection of drone flight data. This comprehensive use of multiple data features can more comprehensively reflect the operating status of the drone and improve the comprehensiveness and accuracy of anomaly detection.

[0047] In summary, this technical solution effectively improves the accuracy and reliability of UAV attitude control anomaly detection by integrating complementary filtering model and LSTM model, dynamically adjusting weights, providing confidence intervals, and comprehensively utilizing multiple data features. It has important practical significance for improving the safety and reliability of UAV flight.

[0048] Based on the above-mentioned data-driven UAV attitude control anomaly detection method, the corresponding application also provides a data-driven UAV attitude control anomaly detection system, which is characterized by comprising: A prediction module is used to train a complementary filter model and an LSTM model based on the sensor data and flight attitude data of the quadrotor drone in a historical time period, and the trained complementary filter model and LSTM model output the first attitude data and the second attitude data; The attitude fusion module is used to determine the attitude data error based on the attitude data at the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment, and to perform weighted fusion on the first attitude data and the second attitude data at the current moment according to the attitude data error to obtain the drone attitude prediction data; The diagnosis module is used to determine the confidence interval according to the residual between the UAV attitude prediction data and the real UAV attitude data in the historical data, and detect the attitude of the UAV according to the confidence interval.

[0049] Based on the above data-driven drone attitude control anomaly detection method, the present application also provides a "cloud-network-terminal" drone cluster control system, including a Kubernetes cluster and a ground station.

[0050] The Kubernetes cluster includes multiple drones, each of which serves as a working node. The ground station serves as the master node to control the working nodes. The above data-driven drone attitude control anomaly detection method is deployed on each working node.

[0051] In this drone cluster control system, the ground station uses Kubernetes cluster technology to achieve centralized management and scheduling of the entire drone cluster. Each drone is added to the Kubernetes cluster as an independent node in the cluster. As the master node, the ground station is responsible for issuing instructions to each drone, monitoring its flight status, and obtaining real-time flight data. Kubernetes's automated deployment and expansion capabilities enable the drone cluster to be flexibly adjusted under changing mission requirements.

[0052] The Kubernetes cluster is constructed as follows: The IP address of each drone in the drone cluster is used as the working node, and the master node is the control end of the drone cluster, which is used to control the flight status of each drone.

[0053] Define the configuration file and set the cloud network drone master node. This node runs on the ground station and is responsible for remotely controlling all drone devices under the drone cloud network platform. Initialize the master node and start the corresponding processes such as etcd (responsible for storing the database of the entire node status and cluster configuration), ApiServer (the only entry for resource operations, receiving user input commands, providing authentication, authorization, and API registration. Other modules query or modify data through apiserver, and only through apiserver can they interact directly with etcd).

[0054] Set the IP address of the node corresponding to the drone so that the IP address of the node in docker can be accessed through the API service interface of the master node, and ensure that there is network isolation in the docker containers of different nodes to ensure resource isolation and fault isolation.

[0055] Use the kubeadm command to add the working node to the k8s overall cluster (a K8s cluster is a collection of physical or virtual machines that are organized into a single computing resource pool and run the Kubernetes platform on it. A K8s cluster usually includes a master node and multiple worker nodes. The master node is usually responsible for cluster management and control, while the worker node is responsible for running containerized applications. That is, the worker node is the corresponding drone device), and start the kubelet and kube-proxy services to forward requests to the pod based on the IP address and port.

[0056] When the model is deployed on the drone node of the cloud-network-end, the actual attitude data of the drone in flight state is calculated with the attitude data output by the model for residual calculation. When the obtained residual data is outside the confidence interval calculated by the sample data, it is judged that the attitude data of the drone at the current time point is abnormal. When the drone flight ends, the abnormal points and abnormal sequences under the drone flight state can be obtained, as follows: During the flight, the drone inputs the drone's sensor data into the trained complementary filtering model and the trained LSTM model respectively. The complementary filtering model outputs the first posture data, and the LSTM model outputs the second posture data. The first posture data and the second posture data are weightedly fused based on the error to obtain the drone's posture prediction data at the current moment, and the residual between the drone's posture prediction data and the actual posture data is calculated, that is, the difference between the actual data and the model prediction data.

[0057] In order to effectively identify anomalies, the calculation results of the residuals are compared with the confidence intervals obtained through sample data analysis. When the residuals exceed the preset confidence interval, the system will immediately determine that the attitude data at the current time point is abnormal. Furthermore, the system can not only mark the anomalies at a single time point, but also track and record abnormal sequences, that is, abnormal data within multiple consecutive time points. The detection of these abnormal points and abnormal sequences helps to detect problems in real time during the flight, such as sensor failures, abnormalities in the flight control system, or external interference. After the flight, all abnormal data points and abnormal sequences will be extracted and analyzed to provide valuable information for subsequent flight data review and troubleshooting, thereby improving the safety and reliability of the drone flight system.

[0058] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each module is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The module described as a separate component may or may not be physically separated. The component displayed as a module may be a physical unit or multiple physical units, that is, it may be located in one place, or it may be distributed in multiple different places. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0059] In addition, each module in each embodiment of the present invention may be integrated into a processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0060] An electronic device provided in an embodiment of the present application includes a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of the data-driven drone attitude control anomaly detection method described in any of the above embodiments are implemented.

[0061] Another electronic device provided in the embodiment of the present application may also include: an input port connected to the processor, used to transmit multimodal data collected by an external acquisition device to the processor; and a display unit connected to the processor, used to display the processing results of the processor to the outside world; a communication module connected to the processor, used to realize the communication between the electronic device and the outside world. The display unit can be a display panel, a laser scanning display, etc.; the communication mode adopted by the communication module includes but is not limited to mobile high-definition link technology (HML), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless connection (including wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, and communication technology based on IEEE802.11s).

[0062] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the data-driven drone attitude control anomaly detection method as described in any of the above embodiments are implemented.

[0063] For the description of the relevant parts of the data-driven unmanned aerial vehicle attitude control anomaly detection system, electronic device, and computer-readable storage medium provided in the embodiments of the present application, please refer to the detailed description of the corresponding parts of the data-driven unmanned aerial vehicle attitude control anomaly detection method provided in the embodiments of the present application, which will not be repeated here. In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0064] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A data-driven method for detecting anomalies in attitude control of unmanned aerial vehicles, characterized in that: include: Based on the sensor data and flight attitude data of the quad-rotor drone in the historical time period, a complementary filter model and an LSTM model are trained respectively, and the trained complementary filter model and LSTM model output the first attitude data and the second attitude data; The attitude data error is determined based on the attitude data of the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment. The first attitude data and the second attitude data at the current moment are weightedly fused according to the attitude data error to obtain the drone attitude prediction data. The confidence interval is determined according to the residual between the predicted UAV attitude data and the actual UAV attitude data in the historical data, and the attitude of the UAV is detected according to the confidence interval.

2. The data-driven unmanned aerial vehicle attitude control anomaly detection method according to claim 1 is characterized in that: The sensor data includes three-axis accelerometer data, three-axis gyroscope data and three-axis magnetometer data; The flight attitude data includes a pitch angle, a roll angle and a yaw angle.

3. The data-driven unmanned aerial vehicle attitude control anomaly detection method according to claim 1 is characterized in that: The complementary filtering model outputs the first posture data including: The sensor data in the navigation coordinate system is converted into the sensor data in the carrier coordinate system, the quantity product error of the sensor data generated in the process of converting the navigation coordinate system and the carrier coordinate system is eliminated, and the sensor data in the carrier coordinate system is converted into Euler angles to obtain the first posture data.

4. The data-driven unmanned aerial vehicle attitude control anomaly detection method according to claim 3 is characterized in that: The sensor data conversion method in the carrier coordinate system is as follows: The quaternion transformation matrix between the navigation coordinate system and the carrier coordinate system is determined, and the sensor data in the navigation coordinate system at each moment is converted into the sensor data in the carrier coordinate system using the quaternion transformation matrix.

5. A data-driven method for detecting abnormality in the attitude control of unmanned aerial vehicles according to claim 3 or 4, characterized in that: The method for determining the first posture data comprises: A PI controller is used to eliminate the vector product error generated in the sensor data conversion process, complete the compensation correction of the sensor data, and then convert the obtained sensor data into the Euler angle form to obtain the first posture data.

6. A data-driven drone attitude control anomaly detection method according to any one of claims 1 to 4, characterized in that: The training method of the LSTM model includes: The sensor data of the drone at the previous n moments and the corresponding posture data are input into the LSTM model as training data, and the LSTM model is sequenced in combination with the stochastic gradient descent method. The trained LSTM model outputs the second posture data.

7. The data-driven method for detecting abnormality in the attitude control of unmanned aerial vehicles according to claim 6, characterized in that: The weighted fusion method of the first posture data and the second posture data is: in, It is the error between the UAV attitude data calculated by the complementary filtering prediction model at the previous moment and the actual attitude data; It is the error between the drone attitude data output by the LSTM prediction model at the last moment and the actual attitude data; is the weight coefficient of the complementary filtering model; is the weight coefficient of the LSTM model; Predict data for drone attitude; is the first posture data; Refers to the second posture data.

8. The data-driven method for detecting abnormality in the attitude control of unmanned aerial vehicles according to claim 1, characterized in that: The detecting the attitude of the UAV according to the confidence interval includes: The sensor data of the UAV are respectively input into the trained complementary filtering model and the trained LSTM model. The complementary filtering model outputs the first attitude data, and the LSTM model outputs the second attitude data. The first attitude data and the second attitude data are weightedly fused based on the error to obtain the UAV attitude prediction data at the current moment. The residual of the UAV attitude prediction data and the actual attitude data is calculated, and the residual is compared with the confidence interval. When the residual exceeds the confidence interval, the attitude data at the current time point is abnormal.

9. A data-driven unmanned aerial vehicle attitude control anomaly detection system, characterized in that: include: A prediction module is used to train a complementary filter model and an LSTM model based on the sensor data and flight attitude data of the quadrotor drone in a historical time period, and the trained complementary filter model and LSTM model output the first attitude data and the second attitude data; The attitude fusion module is used to determine the attitude data error based on the attitude data at the previous moment output by the LSTM model and the actual attitude data of the drone at the corresponding moment, and to perform weighted fusion on the first attitude data and the second attitude data at the current moment according to the attitude data error to obtain the drone attitude prediction data; The diagnosis module is used to determine the confidence interval according to the residual between the UAV attitude prediction data and the real UAV attitude data in the historical data, and detect the attitude of the UAV according to the confidence interval.

10. A "cloud-network-terminal" drone cluster control system, characterized in that: Includes Kubernetes cluster and ground station; The Kubernetes cluster includes multiple drones, each drone serves as a working node, and the ground station serves as a master node to control the working node. The above-mentioned data-driven drone attitude control anomaly detection method is deployed on each working node.

Images