UAV flight data anomaly detection method, device, equipment and storage medium

Abstract

The present invention discloses a method, apparatus, device, and storage medium for detecting anomalies in drone flight data. These methods relate to the field of drone device anomaly detection and address the low accuracy of anomaly detection in drone flight data in the prior art. The method comprises: obtaining historical flight data of a drone, wherein the historical flight data includes multiple flight parameters; inputting the historical flight data into an anomaly detection model for detection, and obtaining detection results corresponding to the multiple flight parameters. The anomaly detection model is constructed by connecting the output of a pretrained denoising autoencoder and the input of a pretrained long-short-term memory neural network. The present invention combines the denoising autoencoder and the long-short-term memory neural network for anomaly detection, thereby improving the accuracy of anomaly detection.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) equipment anomaly detection, and more specifically, to a method, apparatus, device, and storage medium for detecting anomaly in UAV flight data. Background Art

[0002] Accurate and timely drone awareness of its health status is a prerequisite for safe and reliable flight and the foundation for future fully autonomous flight. Flight data is a series of flight parameters related to flight status, and abnormalities often indicate operational risks for the aircraft. Flight data anomaly detection technology facilitates comprehensive monitoring and assessment of drone operational status. Furthermore, real-time anomaly detection and processing of drone flight data enables timely adjustments when flight status anomalies occur, preventing accidents.

[0003] Currently, methods for detecting anomalies in drone flight data primarily include expert knowledge-based, model-based, and data-driven approaches. Expert knowledge-based approaches require domain-level expertise, simulate expert thinking, and employ expert skills. However, their limitations include the difficulty of encoding expert knowledge, poor portability, and poor learning capabilities. Model-based approaches require the creation of a precise drone model. Under the same flight conditions as the drone, the residuals between measurements are calculated as a criterion for determining anomalies. However, establishing precise physical models for each drone subsystem is difficult. Data-driven approaches can be categorized as similarity-based, statistical, classification-based, and prediction-based. However, these methods are all shallow neural learning algorithms, making it difficult to achieve high-accuracy predictions given the temporal and spatial correlation, large volume, high-dimensional multimodality, and high correlation characteristics of drone data.

[0004] In summary, how to solve the low detection accuracy of anomaly detection in UAV flight data is an urgent problem that needs to be solved. Summary of the Invention

[0005] In order to solve the problem of low detection accuracy of anomaly detection of drone flight data in the prior art, the present invention provides a drone flight data anomaly detection method, device, equipment and storage medium. The anomaly detection model trained based on the denoising autoencoder and the long short-term memory neural network can predict output data based on the input historical flight data. The anomaly detection model of the present invention can not only effectively suppress the interference of noise data on anomaly data detection, but also can well explore the spatiotemporal correlation between input data and output results, and provide accurate output parameter estimates. Therefore, the present invention uses a combination of denoising autoencoder and long short-term memory neural network for anomaly detection, thereby improving the accuracy of anomaly detection.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In a first aspect of the present application, a method for detecting anomalies in drone flight data is provided, the method comprising:

[0008] Obtain historical flight data of the UAV, wherein the historical flight data includes multiple flight parameters;

[0009] Historical flight data is input into an anomaly detection model for detection to obtain detection results corresponding to multiple flight parameters. The anomaly detection model is composed of the output end of a pre-trained denoising autoencoder and the input end of a pre-trained long short-term memory neural network.

[0010] In one embodiment, the plurality of flight parameters include the following flight parameters: yaw rate, pitch rate, roll rate, yaw rate, lateral G-load, normal G-load, celestial speed, northing speed, calibrated airspeed, true airspeed, ground speed, axial G-load, pitch angle, roll angle, right inside elevator command, left inside elevator command, right aileron command, left aileron command, altitude, coordinate value vote value, pressure altitude, northing speed, easting speed, and celestial speed.

[0011] In one embodiment, the training process of the denoising autoencoder is as follows: normal flight data without noise is injected with noise of varying degrees to train the denoising autoencoder, so as to obtain a trained denoising autoencoder.

[0012] In one embodiment, the training process of the long short-term memory neural network is: using a trained denoising autoencoder to extract feature vectors of normal flight data, and using the feature vectors to train the long short-term memory neural network to obtain a trained long short-term memory neural network.

[0013] In one embodiment, before inputting the historical flight data into the anomaly detection model for detection, the method further includes:

[0014] Performing standardization processing on historical flight data to obtain standardized historical flight data;

[0015] The standardized historical flight data are cut into fixed time lengths to obtain a flight data set, where the flight data set is a three-dimensional matrix, in which each element includes flight time, memory time length, and multiple flight parameters.

[0016] In one embodiment, historical flight data is input into an anomaly detection model for prediction, and detection results corresponding to multiple flight parameters are obtained, specifically:

[0017] Use the pre-trained denoising autoencoder to extract the flight data of the flight dataset in the initial time series;

[0018] The flight data in the initial time series is processed by sliding windows using a pre-trained long short-term memory neural network, dividing the flight data in the initial time series into multiple time series with fixed window lengths and step sizes. Each time point in the time series corresponds to multiple flight parameters.

[0019] Discrete integration is performed on multiple flight parameters corresponding to each time point in the window to obtain multiple discrete results. When the discrete result is greater than or equal to the standard threshold, the flight parameter at the time point corresponding to the discrete result is marked as abnormal data; the standard threshold is equal to the number of time points in the window multiplied by the time point threshold, and the time point threshold is equal to the sum of the mean of the residual sequence and twice the standard deviation.

[0020] In one embodiment, when the standard deviation of the flight parameter at a time point of the detected data sequence is greater than twice the mean, the data sequence obtained by adding or subtracting the step size of the fixed window from the historical flight data at that time point is marked as abnormal data.

[0021] In a second aspect of the present application, a device for detecting abnormalities in UAV flight data is provided, the device comprising:

[0022] A data acquisition module is used to acquire historical flight data of the UAV, wherein the historical flight data includes multiple flight parameters;

[0023] The anomaly detection module is used to input historical flight data into an anomaly detection model for detection to obtain detection results corresponding to multiple flight parameters. The anomaly detection model is composed of the output end of a pre-trained denoising autoencoder and the input end of a pre-trained long short-term memory neural network.

[0024] The third aspect of the present application provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of a method for detecting anomalies in drone flight data as described in any one of the first aspects of the present application are implemented.

[0025] In a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for detecting anomalies in drone flight data as described in any one of the first aspects of the present application are implemented.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. The drone flight data anomaly detection method provided in this application is based on an anomaly detection model trained by a denoising autoencoder and a long short-term memory neural network. It can predict output data based on the input historical flight data. The anomaly detection model of the present invention can not only effectively suppress the interference of noise data on anomaly data detection, but also can well explore the spatiotemporal correlation between input data and output results, and provide accurate output parameter estimates. Therefore, the present invention uses a combination of a denoising autoencoder and a long short-term memory neural network for anomaly detection, thereby improving the accuracy of anomaly detection.

[0028] 2. The drone flight data anomaly detection method provided in this application also takes into account special anomalies that cannot be detected by discrete integration. For example, the standard deviation of a small number of points in the predicted sequence and the corresponding points in the actual sequence is too large, resulting in the inability of discrete integration to detect anomalies. Therefore, point anomaly capture is used. If the standard deviation of the point is greater than twice the mean, the data sequences before and after the point are marked as abnormal data, thereby further improving the detection accuracy of anomaly detection.

[0029] In addition, the drone flight data anomaly detection devices, equipment and storage media provided in the second to fourth aspects of this application have technical effects corresponding to the above-mentioned drone flight data anomaly detection method, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0031] Figure 1 A flowchart of a method for detecting anomalies in drone flight data provided by an embodiment of the present application;

[0032] Figure 2 This is a structural block diagram of a drone flight data anomaly detection device provided in an embodiment of the present application. Implementation Method

[0033] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0034] As described in the background, current methods for detecting anomalies in drone flight data primarily include expert knowledge-based, model-based, and data-driven approaches. Expert knowledge-based approaches require domain-level expertise, simulate expert thinking, and employ expert skills. However, their limitations include the difficulty of encoding expert knowledge, poor portability, and poor learning capabilities. Model-based approaches require the creation of a precise drone model. Under the same flight conditions as the drone, the residuals between the measured values ​​are calculated as a criterion for determining anomalies. However, establishing precise physical models for each drone subsystem is difficult. Data-driven approaches can be categorized as similarity-based, statistical, classification-based, and prediction-based. However, these methods are all shallow neural learning algorithms, making it difficult to achieve high-accuracy predictions given the temporal and spatial correlation, large volume, high-dimensional multimodality, and high correlation characteristics of drone data. Therefore, in order to solve the problem of low detection accuracy of anomaly detection of drone flight data, the anomaly detection model trained based on the denoising autoencoder and the long short-term memory neural network in the present invention can predict the output data based on the input historical flight data. The anomaly detection model of the present invention can not only effectively suppress the interference of noise data on anomaly data detection, but also can well explore the spatiotemporal correlation between input data and output results, and provide accurate output parameter estimates. Therefore, the present invention uses a combination of denoising autoencoder and long short-term memory neural network for anomaly detection, thereby improving the accuracy of anomaly detection.

[0035] Below, please refer to Figure 1 , Figure 1 A flowchart of a method for detecting abnormalities in drone flight data provided by an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes the following steps:

[0036] S110, obtaining historical flight data of the UAV, where the historical flight data includes multiple flight parameters.

[0037] Specifically, flight data refers to the values ​​of flight parameters related to the performance and flight status of the UAV over a period of time. These values ​​are collected by a large number of sensors installed on the UAV and transmitted to the flight data storage device via the aircraft data bus or to the ground control station via a data link. The collected flight data is divided into power system data, navigation and flight control data, electrical system data, mission equipment data, measurement and control system data, and position information system data according to the signal source. As a specific embodiment, the multiple flight parameters include the following flight parameters: yaw rate, pitch rate, roll rate, yaw rate, lateral overload, normal overload, celestial speed, northing speed, calibrated airspeed, true airspeed, ground speed, axial overload, pitch angle, roll angle, right inner elevator command, left inner elevator command, right aileron command, left aileron command, altitude, coordinate value voting value, barometric altitude, northing speed, easting speed, and celestial speed. It should be understood that the above flight parameters are common knowledge known to those skilled in the art during data anomaly detection. It should be noted that the flight data of the drone is time-related. This is because the sensors arranged on the drone will continuously collect data on various flight parameters when the drone is flying, and these data are related to time.

[0038] Since the sampling frequencies of different sensors are different, flight data needs to be resampled to reasonably supplement the data.

[0039] Since anomaly detection is performed on the drone's flight data after completing a mission, the data obtained is the drone's normal flight status within a certain time period. Since the time period is historical, this is historical flight data. Because sensors collect data on different flight parameters at the same time point within a historical time period, the time points within a historical time period can be divided based on the sensor sampling frequency.

[0040] S120, inputting historical flight data into an anomaly detection model for prediction to obtain anomaly prediction results corresponding to multiple flight parameters, wherein the anomaly detection model is composed of a pre-trained denoising autoencoder and a pre-trained long short-term memory neural network.

[0041] Specifically, the sensor data generated by drones during flight is a typical time series. Prediction-based anomaly detection methods can calculate the current flight status in advance based on historical information. If the predicted data deviates significantly from the actual sensor data, the drone may be under attack and experiencing an anomaly. Long-short-term memory (LSTM) neural networks have a strong approximation capability for nonlinear and non-stationary time series. Therefore, anomaly detection models trained using LSTM neural networks can predict output data based on input data, effectively exploiting the spatiotemporal correlation between input and output, and provide accurate output parameter estimates, thereby ensuring a certain level of detection accuracy.

[0042] The anomaly detection model is obtained by training the original time series of historical flight data based on the long short-term memory neural network. This is a conventional technical means of model training. For example, the original time series data of the historical flight data is divided into a training sample set and a test sample set, and an anomaly detection model is constructed based on the training sample set and the test sample set. The parameters of the anomaly detection model are optimized using the grid search method, and the optimized parameters are substituted into the anomaly detection model to obtain the optimal anomaly detection model; the samples in the test sample set are input into the optimal anomaly detection model to obtain an estimated value sequence; and the residuals between the actual values ​​of the samples in the training sample set and the estimated value sequence are calculated to obtain a residual sequence.

[0043] In one embodiment, the training process of the denoising autoencoder is as follows: normal flight data without noise is injected with noise of varying degrees to train the denoising autoencoder, so as to obtain a trained denoising autoencoder.

[0044] Specifically, the denoising autoencoder is trained by injecting varying degrees of noise into normal, noise-free data. The goal is to learn how to recover the original data from the noisy data. During training, the denoising autoencoder attempts to minimize the reconstruction error, which is the difference between the input data and the data reconstructed by the denoising autoencoder.

[0045] In one embodiment, the training process of the long short-term memory neural network is: using a trained denoising autoencoder to extract feature vectors of normal flight data, and using the feature vectors to train the long short-term memory neural network to obtain a trained long short-term memory neural network.

[0046] Specifically, after training, the denoising autoencoder can be used to denoise the input data and extract features. Therefore, for each time step of the flight data, the trained denoising autoencoder is used to denoise and extract features from the original time series data. The encoding vector and the decoder output are used to represent the original time series data. Furthermore, the extracted feature vectors are used to train a long short-term memory neural network (LSTM) for neural network training, resulting in a trained long short-term memory neural network (LSTM).

[0047] Before inputting the historical flight data into the anomaly detection model for detection, the method also includes: standardizing the historical flight data to obtain standardized historical flight data; cutting the standardized historical flight data according to a fixed time length to obtain a flight data set, wherein the flight data set is a three-dimensional matrix, and each element in the three-dimensional matrix includes flight time, memory time length and multiple flight parameters.

[0048] Specifically, because received drone sensor data often contains missing data and noise, preprocessing of historical flight data is necessary to ensure the accuracy of predictions by the anomaly detection model built using a denoising autoencoder and long-short-term memory neural network. Considering the influence of communication links and transceiver equipment, drone sensor data may be contaminated with noise during generation, transmission, and acquisition, resulting in random deviations within a small range. Due to the high maneuverability of drones, a denoising autoencoder is used to reduce data noise, thereby accurately representing the flight parameters of the drone during its mission. During transmission, due to link unreliability, signals may be unable to be parsed or the parity field of the transmitted data may be corrupted, resulting in ineffective acquisition of drone sensor data and missing data. Therefore, drone sensor data with significant missing data should not be selected as training samples, as this will reduce the accuracy of the long-short-term memory neural network predictions and, in turn, the detection accuracy of the anomaly detection model. Since different sensors have different sampling frequencies, the sampled data needs to be resampled, and the historical flight data needs to be standardized based on the resampling. For example, data normalization can be used for standardization to eliminate the adverse effects of dimensionality between flight parameter indicators. At the same time, it can shorten the training time of the model and make the training process converge as quickly as possible.

[0049] Furthermore, the standardized historical flight data is segmented according to fixed time lengths to obtain a three-dimensional matrix dataset, where each element in the three-dimensional matrix includes the flight time, the memory time length, and the flight parameters collected by the drone sensor. This further captures the temporal correlation of the historical flight data to improve the accuracy of the prediction results output by the subsequent anomaly detection model. In a specific embodiment, the standardized historical flight data can be filtered, where the filtering process specifically includes: when the value of any flight parameter in the historical flight data is infinite, the value of the flight parameter is set to zero. Specifically, as is common knowledge among those skilled in the art, when the value of any flight parameter is infinite, the flight parameter needs to be filtered out. For example, if the yaw angular rate in the flight parameter is infinite, it indicates that the data collected by the sensor is erroneous. In this case, the data needs to be set to 0 to prevent the accuracy of subsequent anomaly detection from being deviated due to obviously erroneous data, thereby improving the detection accuracy of the subsequent anomaly detection model.

[0050] In one embodiment, historical flight data is input into an anomaly detection model for prediction to obtain detection results corresponding to multiple flight parameters, specifically: using a pre-trained denoising autoencoder to extract flight data from the flight data set in an initial time series; using a pre-trained long short-term memory neural network to perform sliding window processing on the flight data in the initial time series, and dividing the flight data in the initial time series into multiple time series with fixed window length and step size; wherein each time point in the time series corresponds to multiple flight parameters; discrete integration is performed on the multiple flight parameters corresponding to each time point in the window to obtain multiple discrete results, and when the discrete result is greater than or equal to a standard threshold, the flight parameter at the time point corresponding to the discrete result is marked as abnormal data; wherein the standard threshold is equal to the number of time points in the window multiplied by the time point threshold, and the time point threshold is equal to the sum of the mean of the residual sequence and twice the standard deviation.

[0051] Specifically, the sliding window processing adopts a fixed window length (c1) and a step size (0.2*c1), where c1 represents the length of the fixed window, and the data points inside the window are discretely integrated. If the calculation results are (ys1+ys2+ys3+…+ysn)*(Δt)≥( +2* )*Δt*n, where is the mean of the residual sequence, is the standard deviation, n is the amount of flight parameter data in a fixed window, y is the absolute value of the difference between the predicted data and the actual (historical flight data), s1, s2, s3…sn are the time points separated by Δt in the fixed window, Δt is the sampling time of the flight parameter, and it means that the flight data of this time series is abnormal.

[0052] The process of determining the residual sequence is as follows: based on the anomaly detection model, normal flight data and historical flight data are respectively detected to obtain a true value sequence and an estimated value sequence; the residual between the estimated value sequence and the true value sequence is calculated to obtain a residual sequence. In this embodiment, the acquired historical flight data is input into the trained anomaly detection model for prediction, thereby obtaining an estimated value sequence, and then the residual sequence is obtained based on the residual between the estimated value sequence and the true value sequence. This is a conventional technical means for those skilled in the art, so no redundant explanation is given on how to calculate the residual sequence. As a preferred embodiment, the allowable range of the residual is ,in, is the mean of the residual sequence, is the standard deviation of the residual sequence, which is the existing technology and should be , where A is the setting coefficient; for example, A can be set to any value from 2 to 5. In this embodiment, it is set to 3. The normal parameter range is .

[0053] In one embodiment, when it is detected that the standard deviation of the flight parameter at a time point of the data sequence is greater than twice the mean, the data sequence obtained by adding or subtracting the step size of the fixed window from the historical flight data at that time point is marked as abnormal data.

[0054] Specifically, when the standard deviation of the flight parameters at a few time points in the data sequence is too large compared to the corresponding points in the true value sequence, the discrete integral cannot detect anomalies, and point anomaly capture is required. If the standard deviation of the flight parameters at this time point is greater than 2 ( is the mean of the residual sequence), then the data sequences before and after this time point are marked as abnormal data. Therefore, this embodiment takes into account special abnormal situations that cannot be detected by discrete integration, further improving the detection accuracy of abnormality detection.

[0055] Corresponding to the above-mentioned embodiment of the method for detecting abnormality of UAV flight data, the embodiment of the present invention further provides a device for detecting abnormality of UAV flight data, such as Figure 2 As shown, the device includes:

[0056] A data acquisition module 210 is used to acquire historical flight data of the UAV, wherein the historical flight data includes multiple flight parameters;

[0057] Anomaly detection module 220 is used to input historical flight data into an anomaly detection model for detection to obtain detection results corresponding to multiple flight parameters, wherein the anomaly detection model is composed of the output end of a pre-trained denoising autoencoder and the input end of a pre-trained long short-term memory neural network.

[0058] In summary, the drone flight data anomaly detection device provided by the embodiment of the present invention has the following beneficial effects:

[0059] 1. The drone flight data anomaly detection device provided in this application is based on an anomaly detection model trained by a denoising autoencoder and a long short-term memory neural network. It can predict output data based on the input historical flight data. The anomaly detection model of the present invention can not only effectively suppress the interference of noise data on anomaly data detection, but also can well explore the spatiotemporal correlation between input data and output results, and provide accurate output parameter estimates. Therefore, the present invention uses a combination of a denoising autoencoder and a long short-term memory neural network for anomaly detection, thereby improving the accuracy of anomaly detection.

[0060] 2. The drone flight data anomaly detection device provided in this application also takes into account special anomalies that cannot be detected by discrete integration. For example, the standard deviation between a small number of points in the predicted sequence and the corresponding points in the actual sequence is too large, resulting in the inability of discrete integration to detect anomalies. Therefore, point anomaly capture is used. If the standard deviation of the point is greater than twice the mean, the data sequences before and after the point are marked as abnormal data, thereby further improving the detection accuracy of anomaly detection.

[0061] In another embodiment of the present invention, an electronic device is provided, comprising one or more processors; a memory coupled to the processors for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the drone flight data anomaly detection method described in the above embodiment. The processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used to perform the operations of the drone flight data anomaly detection method.

[0062] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It is understood that the computer-readable storage medium herein may include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be high-speed RAM memory or non-volatile memory, such as at least one disk drive. The processor may load and execute the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for detecting anomalies in drone flight data described in the above-mentioned embodiment. Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0063] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting anomalies in UAV flight data, characterized in that: Methods include: Obtain historical flight data of the UAV, wherein the historical flight data includes multiple flight parameters; Input historical flight data into the anomaly detection model for prediction, and obtain detection results corresponding to multiple flight parameters. The anomaly detection model is composed of the output end of a pre-trained denoising autoencoder and the input end of a pre-trained long short-term memory neural network. Input historical flight data into the anomaly detection model for prediction, and obtain detection results corresponding to multiple flight parameters, specifically: Use the pre-trained denoising autoencoder to extract the flight data of the flight dataset in the initial time series; The flight data in the initial time series is processed by sliding windows using a pre-trained long short-term memory neural network, dividing the flight data in the initial time series into multiple time series with fixed window lengths and step sizes. Each time point in the time series corresponds to multiple flight parameters. Perform discrete integration on multiple flight parameters corresponding to each time point within the window to obtain multiple discrete results. When the discrete results are greater than or equal to the standard threshold, the flight parameters at the time point corresponding to the discrete results are marked as abnormal data. The standard threshold is equal to the number of time points in the window multiplied by the time point threshold, and the time point threshold is equal to the sum of the mean and twice the standard deviation of the residual series.

2. The method for detecting anomalies in UAV flight data according to claim 1, wherein: The multiple flight parameters include the following flight parameters: yaw rate, pitch rate, roll rate, yaw rate, lateral G, normal G, celestial speed, northing speed, calibrated airspeed, true airspeed, ground speed, axial G, pitch angle, roll angle, right inner elevator command, left inner elevator command, right aileron command, left aileron command, altitude, coordinate value voting value, pressure altitude, northing speed, easting speed and celestial speed.

3. The method for detecting anomalies in UAV flight data according to claim 1, wherein: The training process of the denoising autoencoder is as follows: normal flight data without noise is injected with noise of varying degrees to train the denoising autoencoder, so as to obtain a trained denoising autoencoder.

4. The method for detecting anomalies in UAV flight data according to claim 3, wherein: The training process of the long short-term memory neural network is as follows: using the trained denoising autoencoder to extract the feature vector of normal flight data, and using the feature vector to train the long short-term memory neural network to obtain a trained long short-term memory neural network.

5. The method for detecting anomalies in drone flight data according to claim 1, wherein: Before historical flight data is fed into the anomaly detection model for prediction, it also includes: Performing standardization processing on historical flight data to obtain standardized historical flight data; The standardized historical flight data are cut into fixed time lengths to obtain a flight data set, where the flight data set is a three-dimensional matrix, in which each element includes flight time, memory time length, and multiple flight parameters.

6. The method for detecting anomalies in UAV flight data according to claim 1, wherein: When the standard deviation of the flight parameters at a time point of the data sequence is detected to be greater than twice the mean, the data sequence obtained by adding or subtracting the step size of the fixed window from the historical flight data at that time point is marked as abnormal data.

7. A device for detecting abnormalities in UAV flight data, characterized in that: The device includes: A data acquisition module is used to acquire historical flight data of the UAV, wherein the historical flight data includes multiple flight parameters; The anomaly detection module is used to input historical flight data into an anomaly detection model for prediction, and obtain detection results corresponding to multiple flight parameters. The anomaly detection model is composed of the output end of a pre-trained denoising autoencoder and the input end of a pre-trained long short-term memory neural network. The historical flight data is input into the anomaly detection model for prediction, and the detection results corresponding to multiple flight parameters are obtained, specifically: Use the pre-trained denoising autoencoder to extract the flight data of the flight dataset in the initial time series; The flight data in the initial time series is processed by sliding windows using a pre-trained long short-term memory neural network, dividing the flight data in the initial time series into multiple time series with fixed window lengths and step sizes. Each time point in the time series corresponds to multiple flight parameters. Perform discrete integration on multiple flight parameters corresponding to each time point within the window to obtain multiple discrete results. When the discrete results are greater than or equal to the standard threshold, the flight parameters at the time point corresponding to the discrete results are marked as abnormal data. The standard threshold is equal to the number of time points in the window multiplied by the time point threshold, and the time point threshold is equal to the sum of the mean and twice the standard deviation of the residual series.

8. An electronic device, characterized in that: The electronic device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the drone flight data anomaly detection method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method for detecting anomalies in flight data of a drone according to any one of claims 1 to 6 are implemented.

Images