High-precision anomaly detection and data recovery method for flight data of unmanned aerial vehicle

Abstract

The invention relates to the technical field of unmanned aerial vehicle detection, in particular to an unmanned aerial vehicle flight data high-precision anomaly detection and data recovery method. Comprising the following steps that S1, a multiple regression model based on TCN-KANs is constructed, the TCN-KANs model comprises an input layer, three residual blocks and a KAN layer, the residual blocks extract features through a convolution layer and residual connection, nonlinear change is carried out on the input features, a potential mode is extracted, final output is generated, and the KAN layer serves as a final output layer of TCN-KANs; s2, minimizing an error between a predicted value and a real value; s3, calculating a residual error according to a prediction result of the model, carrying out smoothing processing on the residual error, and analyzing statistical characteristics of the residual error to determine an abnormal threshold value; and S4, during real-time flight monitoring and operation control of the unmanned aerial vehicle, generating a recovery value for data recovery after abnormality is detected through the model.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle detection, and in particular to a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data. Background Art

[0002] With the continuous expansion of drone application fields and the rapid growth of market scale, people are paying more and more attention to and studying drones. Drones are complex physical closed-loop control systems that face the challenge of lack of real-time observation and rapid decision-making capabilities of pilots. This makes the demand for the safety and reliability of drones more urgent. Drones rely on sensors such as GPS, gyroscopes, and accelerometers to collect information related to the flight status of drones. This information reflects the performance indicators of drones such as flight trajectory, attitude changes, and motion status, providing an important data basis for their health management and operation and maintenance. By conducting in-depth analysis of these data, real-time monitoring and evaluation of the operating status of drones can be achieved, thereby significantly improving the reliability and safety of operations. In this context, anomaly detection of flight data has become one of the important means of drone status monitoring. By processing and analyzing flight data, deviations from normal flight modes can be discovered in a timely manner. Once an anomaly is identified, the flight path or control instructions of the drone can be actively adjusted by further implementing abnormal recovery or mitigation measures to ensure the safe operation of the drone. Therefore, actively carrying out research on anomaly detection and recovery of drone flight data has important practical value and broad application prospects.

[0003] At present, many studies on anomaly detection of drone flight data have been carried out in the existing technology. From the perspective of methodology, they are mainly divided into methods based on traditional machine learning and deep learning, including K-means, SVM, KPCA, KNN, decision tree (DT), random forest (RF), Kalman filter (KF), LSTM, BiLSTM, 1D CNN-LSTM, etc. However, the following problems still exist: 1. Although these traditional machine learning-based methods are widely used for anomaly or fault detection in UAV flight data, they often show limited flexibility and adaptability when dealing with complex high-dimensional data. Secondly, these methods often require manual feature engineering, which is not only time-consuming and labor-intensive, but may also lead to the omission of important features. In addition, for application scenarios that emphasize the integrity and reliability of data after anomaly detection, it is necessary to reasonably recover the detected anomaly data. However, most of the above methods lack analytical redundancy and cannot effectively achieve anomaly data recovery.

[0004] 2. Deep learning-based methods often use LSTM to model flight data, which may lead to gradient vanishing or exploding problems when processing long sequences of flight data, thus affecting the performance of the model. Although LSTM-CNN performs well on short-term dependencies, the fixedness and local characteristics of the convolution kernel limit its ability to process complex high-dimensional and nonlinear time series data, affecting the overall performance of the model. Therefore, although these methods have achieved remarkable results in anomaly and fault detection, there are still many challenges in building accurate prediction models when processing flight data with long-term dependencies and complex time series structures. This requires the model to have not only efficient feature extraction capabilities, but also excellent time series representation capabilities to accurately capture long-term dependencies, thereby significantly improving overall performance.

[0005] 3. Existing deep learning methods usually use the fully connected layer of the multi-layer perceptron (MLP) to further process the extracted features. However, since MLP uses fixed activation functions and linear transformations on nodes, it lacks sufficient flexibility and adaptability in representing complex time series data and modeling nonlinear relationships, resulting in the inability to fully utilize the features extracted by the model, which may reduce model performance. Summary of the invention

[0006] The technical problem solved by the present invention is to provide a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data, which can improve the flight safety and data reliability of the unmanned aerial vehicle.

[0007] The basic scheme provided by the present invention comprises the following steps: A method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data comprises the following steps: S1. Construct a multivariate regression model based on TCN-KANs. The TCN-KANs model includes an input layer, three residual blocks and a KAN layer. KANs is used as the final output layer of TCN-KANs. The residual block extracts features through convolutional layers and residual connections, and performs nonlinear changes on the input features to extract potential patterns and generate final outputs. S2, minimize the error between the predicted value and the true value; S3. Calculate the residual according to the prediction results of the model, smooth the residual, and determine the abnormal threshold by analyzing the statistical characteristics of the residual; S4. During the real-time monitoring and operation control of the UAV flight, the model generates a recovery value for data recovery after an anomaly is detected.

[0008] Further, the S1 comprises the following steps: S11. Set the first residual block to include two extended causal convolutional layers, with a kernel size of 3, an expansion factor of 1, and a number of filters of 16. In the second residual block, the kernel size is 3, the expansion factor is 2, and the number of filters is 32. In the third residual block, the kernel size is 3, the expansion factor is 4, and the number of filters is 64. The output after three residual blocks is expressed as follows:

[0009] in , and Represent the outputs of the first, second, and third residual blocks, respectively. Represents the output features after being processed by the mapping function TCN(⋅); S12. Through the KAN layer, the input features are transformed nonlinearly, the potential patterns are extracted and the final output is generated, as shown in the following formula:

[0010] in The final prediction result is Represents a mapping function.

[0011] Further, the S2 comprises the following steps: S21. Update the weights by calculating the gradient of the network to minimize the predicted value and the true value The error between them is calculated until the predetermined model accuracy requirement is met, and the average error MSE is used as the loss function:

[0012] In the formula and Respectively The true and predicted values ​​of the data points. is the sample length.

[0013] Further, the S3 comprises the following steps: S31. Calculate the residual according to the prediction results of the model and input the training set into the model respectively. and test set ,get:

[0014]

[0015] in, and They are and The predicted value of is the mapping function; S32, training residual and the test residuals :

[0016]

[0017] in and are the training set and test set of the target variable respectively; S33, introduced the exponential weighted average method EWMA method and To perform smoothing:

[0018] in, It is The smoothed value of the residual, is an adjustable weight parameter, is the original residual, and ; S34, SPOT method based on extreme value theory is used to obtain dynamic thresholds, and the part exceeding the threshold is expressed as :

[0019] in, The initial threshold is Initialize, γ and σ are shape parameters and scale parameters respectively; S35. Use the likelihood function maximization method to estimate the γ and σ parameters:

[0020]

[0021] in is greater than The sum of the sample data of the peak value; and is the update parameter obtained by the maximum likelihood estimation method, q represents the predefined extreme point quantile, is the total number of samples. is the final threshold and The difference value of ; S36, according to S35 ; S37. According to S33 and S36, perform anomaly detection:

[0022] in is the smoothed test residual No. samples, yes The threshold value is 1, indicating abnormality and 0, indicating normality.

[0023] Further, the S4 comprises the following steps: S41, the prediction result in S31 As a recovery value, the data deviation is corrected.

[0024] The principles and advantages of the present invention are: 1. A UAV flight data anomaly detection and recovery framework TCKANs-ADR is proposed. The framework integrates advanced deep learning network architecture and anomaly detection strategy, providing a more accurate and robust solution to the challenges of complex time series data modeling, time series feature extraction, and random noise influence.

[0025] 2. A multivariate regression method based on TCN and KANs (TCN-KANs) is proposed. This method inherits the advantages of TCN in processing long sequence data and extracting time series features. At the same time, KANs is introduced into the field of anomaly detection and recovery of drone flight data for the first time, and it cleverly replaces the fully connected layer in TCN, thereby enhancing the modeling ability of complex spatiotemporal correlation flight data and improving the accuracy and robustness of model prediction.

[0026] 3. A dynamic threshold calculation method based on extreme value theory and residual smoothing is proposed. This method first uses the exponential weighted average method to smooth the residual to overcome the interference of random noise. Then, by analyzing the extreme value distribution of the residual historical data, the threshold is dynamically adjusted to adapt to the changes in flight data, thereby improving the model anomaly detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to the present invention; Figure 2 A schematic diagram of the TCN-KANs model structure of an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to the present invention; Figure 3 A schematic diagram of a partial parameter curve of flight data of an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data of the present invention; Figure 4A schematic diagram of flight data after anomaly injection according to an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data of the present invention; Figure 5 A visualization diagram of anomaly detection results of an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data of the present invention; Figure 6 A visualization diagram of data recovery results of an embodiment of a method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data of the present invention; DETAILED DESCRIPTION The following is further described in detail through specific implementation methods: The embodiment is basically as shown in the attached Figure 1 As shown: A method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data, comprising the following steps: S1. Construct a multivariate regression model based on TCN-KANs. The TCN-KANs model includes an input layer, three residual blocks and a KAN layer. KANs is used as the final output layer of TCN-KANs. The residual block extracts features through convolutional layers and residual connections, and performs nonlinear changes on the input features to extract potential patterns and generate the final output.

[0028] Specifically, accurate anomaly detection depends on good model prediction capabilities. TCN has shown superior performance over LSTM and CNN in time series prediction. Thanks to its causal convolution and dilated convolution characteristics, it can effectively capture long-term dependencies and avoid gradient vanishing and explosion problems. In addition, the parallel computing capability of TCN significantly improves the training speed, making it more efficient when processing large-scale data. Although these advantages have made TCN a focus of attention in time series prediction tasks, its multi-layer perceptron (MLP) used as the output layer may not fully exploit the spatiotemporal features extracted by TCN. Due to the high dimensionality, nonlinearity, and dynamic changes of drone flight data, MLP may not be able to effectively capture these complex data features, which may lead to limited improvements in the model's anomaly detection and prediction performance. However, the emergence of KANs provides a possible solution to this problem. Therefore, this application constructs the following Figure 2 The multivariate regression model based on TCN-KANs shown in the figure is designed to more effectively capture the complex dynamic characteristics of UAV flight data and achieve accurate prediction.

[0029] The designed TCN-KANs model consists of an input layer, three residual blocks and a KAN layer. Unlike traditional TCN, the fully connected layer in TCN is abandoned, and KANs is used as the final output layer of TCN-KANs. The residual block effectively extracts features through convolutional layers and residual connections to enhance the model's processing capabilities for time series data. The first residual block consists of two extended causal convolutional layers. The kernel size is set to 3, the expansion factor is set to 1, and the number of filters is set to 16. In the second residual block, the kernel size is set to 3, the expansion factor is set to 2, and the number of filters is set to 32. Finally, in the third residual block, the kernel size is also set to 3, the expansion factor is set to 4, and the number of filters is set to 64. For multivariate flight data , which first performs initial feature extraction by TCN.

[0030] S1 includes the following steps: S11. Set the first residual block to include two extended causal convolutional layers, with a kernel size of 3, an expansion factor of 1, and a number of filters of 16. In the second residual block, the kernel size is 3, the expansion factor is 2, and the number of filters is 32. In the third residual block, the kernel size is 3, the expansion factor is 4, and the number of filters is 64. The output after three residual blocks is expressed as follows:

[0031] in , and Represent the outputs of the first, second, and third residual blocks, respectively. Represents the output features after being processed by the mapping function TCN(⋅); S12. Through the KAN layer, the input features are transformed nonlinearly, the potential patterns are extracted and the final output is generated, as shown in the following formula:

[0032] in The final prediction result is Represents a mapping function.

[0033] S2. Minimize the error between the predicted value and the true value.

[0034] The S2 comprises the following steps: The TCN-KANs model is trained based on the back-propagation algorithm. The algorithm updates the weights by calculating the gradient of the network to minimize the predicted value. and the true value until the predetermined model accuracy requirement is met.

[0035] S21. Update the weights by calculating the gradient of the network to minimize the predicted value and the true value The error between them is calculated until the predetermined model accuracy requirement is met, and the average error MSE is used as the loss function:

[0036] In the formula and Respectively The true and predicted values ​​of the data points. is the sample length.

[0037] This application uses the fast-converging Adam optimizer to update the network weights and combines it with regularization technology to efficiently train the TCN-KANs model, thereby achieving superior performance and high accuracy in real applications.

[0038] S3. Calculate the residual according to the prediction results of the model, smooth the residual, and determine the abnormal threshold by analyzing the statistical characteristics of the residual.

[0039] S3 includes the following steps: S31. Calculate the residual according to the prediction results of the model and input the training set into the model respectively. and test set ,get:

[0040]

[0041] in, and They are and The predicted value of is the mapping function; S32, training residual and the test residuals :

[0042]

[0043] in and are the training set and test set of the target variable respectively; Due to the presence of random noise in drone flight data, the residuals fluctuate violently, which interferes with anomaly detection. Therefore, the exponentially weighted average method (EWMA) is introduced to and Smoothing is performed to suppress the influence of random noise.

[0044] S33, introduced the exponential weighted average method EWMA method and To perform smoothing:

[0045] in, It is The smoothed value of the residual, is an adjustable weight parameter, is the original residual, and ; In addition, in order to adapt the anomaly detection threshold to the performance of the UAV in a dynamically changing flight environment, the SPOT method based on extreme value theory is further introduced to obtain the dynamic threshold. This method effectively avoids the strict assumptions of traditional statistical thresholds on data distribution, such as normality. In addition, the SPOT method has good scalability and can adaptively adjust the threshold under different environmental conditions, thereby improving the anomaly detection capability and flight safety of the UAV. The SPOT method obtains the threshold based on the distribution of extreme values.

[0046] S34, SPOT method based on extreme value theory is used to obtain dynamic thresholds, and the part exceeding the threshold is expressed as :

[0047] in, The initial threshold is Initialize, γ and σ are shape parameters and scale parameters respectively; S35. Use the likelihood function maximization method to estimate the γ and σ parameters:

[0048]

[0049] in is greater than The sum of the sample data of the peak value; and is the update parameter obtained by the maximum likelihood estimation method, q represents the predefined extreme point quantile, is the total number of samples. is the final threshold and The difference value of ; S36, according to S35 ; S37. According to S33 and S36, perform anomaly detection:

[0050] in is the smoothed test residual No. samples, yes The threshold value is 1, indicating abnormality and 0, indicating normality.

[0051] S4. During the real-time monitoring and operation control of the UAV flight, the model generates a recovery value for data recovery after an anomaly is detected.

[0052] In the real-time monitoring and operation control scenarios of UAV flight, effective data recovery can quickly restore the system to normal state after detecting an abnormality, thereby maintaining the stability and reliability of the UAV system. By using the expected output of the model for the normal state, the prediction result in equation (5) is As a recovery value, it can quickly correct data deviations and reduce flight risks caused by abnormalities.

[0053] S4 includes the following steps: S41, the prediction result in S31 As a recovery value, the data deviation is corrected.

[0054] The following steps are also included: S5. Generate experimental data and abnormal data to verify the reliability of the model.

[0055] The experimental data comes from the actual flight data collected autonomously by a real composite wing UAV, which is transmitted to the ground station in real time through a digital radio to achieve real-time recording and storage of data. A total of parameters collected by sensors including barometers, accelerometers, gyroscopes, magnetometers, etc. were collected, including but not limited to airspeed, roll angle, eastward speed, longitude, latitude, pitch angle, altitude, etc. 16 flight parameters related to the attitude control of the UAV were selected as experimental data. The names of these parameters are GPS_Alt, GPS_Lat, GPS_Lon, V_east, V_north, Vz, AirSpeed, Yaw, Pitch, Roll, Acc_X, Acc_Y, Acc_Z, wX, wY and wZ, and their meanings and units are shown in Table 1. The above parameters with a length of 12000 were intercepted as model input.

[0056] Table 1

[0057] Figure 3 Some parameter curves of the flight data used are shown.

[0058] Since it is difficult and costly to obtain actual abnormal data, the anomaly injection method is used to obtain abnormal data. Acc_Z is used as the monitoring parameter, and deviation and drift anomalies are injected into it according to formula (21).

[0059]

[0060] where y(t) and They are normal and abnormal flight data, is a constant, is a function of t. In this paper, Set to 0.1. , we take values ​​at equal intervals in the interval [0.08, 0.12]. For the flight parameter Acc_Z, we injected 2000 abnormal points with an abnormal rate of 0.167. The injection of abnormal points starts from the 10000th sample point and continues until the end of the data. This means that in the entire data sequence, the first 10000 sample points are normal, while the part after the 10000th sample point contains artificially introduced abnormal data. Figure 4 Flight data with original and injected deviations and anomalies are shown.

[0061] In this application, the maximum and minimum method is used to normalize the parameters shown in Table 1. Maximum and minimum normalization is a commonly used data preprocessing technique, which aims to unify parameters of different dimensions and ranges to the same standard for subsequent analysis and to avoid the impact of different dimensions on model performance. Specifically, this method converts each parameter value into a value in the range of [0, 1] using the following formula:

[0062] in, represents normalized data, Represents the original data, and express The maximum and minimum values ​​of .

[0063] This application uses accuracy (Acc), true positive rate (TPR), recall rate (Rec), precision (Pre), false positive rate (FPR) and F1 score as evaluation indicators for anomaly detection, as shown in formulas (16)-(20). Mean absolute error (MAE) and MSE are used as evaluation indicators for data recovery:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] Among them, TP and TN represent the correctly identified normal samples and abnormal samples, FP and FN represent the normal samples and abnormal samples that are misclassified as abnormal samples and normal samples. and Represent the original value and predicted value of the ith sample, respectively, and N represents the total number of samples. The larger the TPR, Acc, Pre and F1 values ​​are, and the smaller the FPR, MAE and MSE values ​​are, the better the performance of the model is.

[0071] To verify the effectiveness of TCKANs-ADR, it is compared with multiple benchmark regression methods, including support vector regression (SVR), LSTM, BiLSTM, 1D CNN-LSTM, and 1D CBiAM. These methods cover advanced regression techniques based on traditional machine learning and deep learning, thus providing a comprehensive comparative perspective. Figure 5 The anomaly detection visualization results of TCKANs-ADR are shown. It can be seen that although TCKANs-ADR has a small amount of FN for different anomaly types, it has a higher TN and less FP for most of the anomaly point pairs. This shows that the model has good performance and stability in distinguishing normal and abnormal data. To quantify the anomaly detection performance of TCKANs-ADR and the above-mentioned baseline methods, Table 2 shows the results of each method in the anomaly detection task. Thanks to the advantages of TCN and KANs, as well as the residual smoothing method and dynamic threshold strategy, TCKANs-ADR achieves the best anomaly detection results compared with the baseline method. For deviation anomalies, the TPR, FPR, Acc, Pre and F1 values ​​of TCKANs-ADR are 90.69%, 1.00%, 95.30%, 98.64% and 94.50% respectively. For drift anomalies, the TPR, FPR, Acc, Pre, and F1 values ​​of TCKANs-ADR are 86.00%, 0.15%, 93.69%, 99.78%, and 92.38%, respectively. These results confirm the stability and reliability of TCKANs-ADR in processing different types of anomaly data.

[0072] Table 2

[0073] The FPR, Acc, Pre, and F1 values ​​of SVR in deviation and drift anomaly detection are 8.85% and 3.90%, 79.31% and 82.06%, 85.36% and 92.97%, and 73.48% and 76.16%, respectively. However, the true TPR value of SVR is only 64.50%, which shows that although the model performs well in distinguishing normal and abnormal samples, a considerable number of true normal samples are still not detected, which may cause potential problems to be ignored. LSTM also performs poorly in the detection of deviation and drift anomalies, with FPR values ​​as high as 56.28% and 63.73%, respectively, and low Acc, Pre, and F1 values. This may be due to the limited ability of LSTM to capture dependencies when processing non-stationary flight data. At the same time, the lack of a suitable noise processing strategy leads to a decrease in the detection accuracy of LSTM. Although BiLSTM also outperforms LSTM in overall performance, its performance in FPR, Acc, Pre, and F1 values ​​is still not ideal. This may still be due to the interference of random noise and the insufficient feature extraction capability of BiLSTM, which affects its performance in anomaly detection.

[0074] 1D CNN-LSTM combines the feature extraction of 1D CNN and the temporal dependency modeling capability of LSTM, and shows relatively good detection performance. In the detection of deviation anomalies, the TPR, Acc, Pre and F1 values ​​of 1D CNN-LSTM are 81.69%, 86.69%, 87.54% and 84.51% respectively, but its FPR value is relatively high, reaching 9.30%. For drift anomalies, although the overall performance of 1D CNN-LSTM has declined, it still maintains high Acc and Pre, which are 80.83% and 90.34% respectively, while the FPR is significantly reduced to 5.45%. Similarly, 1D CBiAM, which combines the advantages of multiple deep learning, also shows good anomaly detection performance overall. Compared with LSTM, BiLSTM and 1D CNN-LSTM, 1D CBiAM achieved the lowest FPR values ​​of 2.90% and 4.85% respectively, and achieved higher Acc, Pre and F1 values. However, 1D CBiAM achieves lower TPR values ​​of 69.69% and 71.56% for bias and drift anomalies, respectively. This indicates that although 1D CBiAM performs well in FPR and other evaluation metrics, it still has significant room for improvement in TPR.

[0075] Tables 3 and 4 list the performance differences between TCKANs-ADR and the baseline methods under two anomaly detection tasks. Whether it is deviation anomaly or drift anomaly, TCKANs-ADR has made significant improvements in multiple performance indicators compared to the baseline method. Specifically, in the deviation anomaly detection task, compared with the baseline method, TCKANs-ADR's performance in TPR, Acc, Pre, and F1 is 9.00% to 26.19%, 8.61% to 39.04%, 3.58% to 48.07%, and 9.98% to 35.11% higher. In addition, TCKANs-ADR is significantly lower than the baseline method in FPR. The FPR ranges from -55.28% to -1.90%. For drift anomaly, TCKANs-ADR outperforms the baseline method by TPR, Acc, Pre and For F1, the increases were 13.81% to 22.31%, 9.03% to 41.46%, 6.81% to 52.23%, and 11.80% to 35.05%, respectively. The FPR is -63.58%~-3.75% lower. These results further demonstrate the effectiveness and superiority of TCKANs-ADR's anomaly detection performance.

[0076] Table 3

[0077] Table 4

[0078] Figure 6 The data recovery visualization results of TCKANs-ADR and the baseline methods are shown. It can be seen that, for example, LSTM, BiLSTM, 1D CNN-LSTM, and 1D CBiAM methods all have obvious recovery deviations for data recovery results of different anomaly types. In contrast, TCKANs-ADR can still fit the data trend and distribution and recover the data more accurately.

[0079] Table 5 lists the recovery results of the above methods. It can be seen that LSTM performs the worst, with MAE and MSE values ​​of 0.041376 and 0.002549 for deviation and 0.052706 and 0.003923 for drift, respectively. This is mainly because complex flight data often have strong nonlinear and time-varying characteristics, and LSTM may be limited by its structure in modeling these features. Compared with LSTM, the MAE and MSE values ​​of BiLSTM and 1D CNN-LSTM for deviation and drift anomalies have improved to a certain extent, but are still relatively high. This shows that these models still face challenges when dealing with flight data with strong nonlinear and time-varying characteristics. Although 1D CBiAM performs well in the recovery of deviation anomalies, 0.033561 and 0.001761, respectively, it has higher MAE and MSE values ​​for drift anomalies, 0.042438 and 0.002674, respectively. SVR has good overall data recovery performance, with MAE and MSE values ​​of 0.032566 and 0.001742 and 0.032548 and 0.001740 for deviation and drift anomalies, respectively. Compared with the baseline method, TCKANs-ADR performs best in handling deviation and drift anomalies, with the lowest MAE and MSE values ​​of 0.031322 and 0.001619 and 0.031317 and 0.001635 for deviation and drift anomalies, respectively. This superior performance is due to TCKANs-ADR's advantages in complex nonlinear relationship modeling and feature extraction, which enables it to better fit the changes in flight data distribution.

[0080] Table 5

[0081] To further explore the data recovery differences between different anomaly types, the difference values ​​of various methods are listed in Table 6. The ∆MAE and ∆MSE of LSTM are still the highest, which are 0.011330 and 0.001374, respectively. This shows that LSTM is most affected by the switching of different anomaly models and cannot effectively adapt to the changes in anomaly patterns, resulting in large fluctuations in its recovery performance under different anomaly conditions. In contrast, the ∆MAE and ∆MSE values ​​of BiLSTM, 1D CNN-LSTM, and 1D CBiAM are relatively low, but are still affected to some extent by the switching of different anomaly patterns. For example, the ∆MAE and ∆MSE values ​​of 1D CNN-LSTM are 0.011152 and 0.001172, respectively. The overall data recovery performance of SVR is good, especially it has the lowest ∆MSE value of 0.000002. Although the ∆MSE value of TCKANs-ADR is slightly lower than that of SVR, which is 0.000016, it has the lowest ∆MAE value of 0.000005. This shows that TCKANs-ADR has high adaptability and stability in switching between different anomaly modes while maintaining the best anomaly detection performance. Overall, TCKANs-ADR can not only effectively identify and recover anomaly data, but also maintain consistent recovery effects when facing different anomaly types, showing stronger robustness.

[0082] Table 6

[0083] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the relevant field know all the common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for high-precision anomaly detection and data recovery of UAV flight data, characterized by: The following steps are involved: S1. Construct a multivariate regression model based on TCN-KANs. The TCN-KANs model includes an input layer, three residual blocks and a KAN layer. KANs is used as the final output layer of TCN-KANs. The residual block extracts features through convolutional layers and residual connections, and performs nonlinear changes on the input features to extract potential patterns and generate final outputs. S2, minimize the error between the predicted value and the true value; S3. Calculate the residual according to the prediction results of the model, smooth the residual, and determine the abnormal threshold by analyzing the statistical characteristics of the residual; S4. During the real-time monitoring and operation control of the UAV flight, the model generates a recovery value for data recovery after an anomaly is detected.

2. The method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to claim 1, characterized in that: The S1 comprises the following steps: S11. Set the first residual block to include two extended causal convolutional layers, with a kernel size of 3, an expansion factor of 1, and a number of filters of 16. In the second residual block, the kernel size is 3, the expansion factor is 2, and the number of filters is 32. In the third residual block, the kernel size is 3, the expansion factor is 4, and the number of filters is 64. The output after three residual blocks is expressed as follows: in , and Represent the outputs of the first, second, and third residual blocks, respectively. Represents the output features after being processed by the mapping function TCN(⋅); S12. Through the KAN layer, the input features are transformed nonlinearly, the potential patterns are extracted and the final output is generated, as shown in the following formula: in Represents the final prediction result. Represents a mapping function.

3. A method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to claim 2, characterized in that: The S2 comprises the following steps: S21. Update the weights by calculating the gradient of the network to minimize the predicted value and the true value The error between them is calculated until the predetermined model accuracy requirement is met, and the average error MSE is used as the loss function: In the formula and Respectively The true and predicted values ​​of the data points. is the sample length.

4. The method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to claim 3 is characterized in that: The S3 comprises the following steps: S31. Calculate the residual according to the prediction results of the model and input the training set into the model respectively. and test set ,get: in, and They are and The predicted value of is the mapping function; S32, training residual and the test residuals : in and are the training set and test set of the target variable respectively; S33, introduced the exponential weighted average method EWMA method and To perform smoothing: in, It is The smoothed value of the residual, is an adjustable weight parameter, is the original residual, and ; S34, based on the extreme value theory, the flow peak value exceeds the threshold SPOT method to obtain the dynamic threshold, and the part exceeding the threshold is expressed as : in, The initial threshold is Initialize, γ and σ are shape parameters and scale parameters respectively; S35. Use the likelihood function maximization method to estimate the γ and σ parameters: in is greater than The sum of the sample data of the peak value; and is the update parameter obtained by the maximum likelihood estimation method, q represents the predefined extreme point quantile, is the total number of samples, is the final threshold and The difference value of ; S36, according to S35 ; S37. According to S33 and S36, perform anomaly detection: in is the smoothed test residual No. samples, yes The threshold value is 1, indicating abnormality and 0, indicating normality.

5. The method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to claim 4 is characterized in that: The S4 comprises the following steps: S41, the prediction result in S31 As a recovery value, the data deviation is corrected.

6. A method for high-precision anomaly detection and data recovery of unmanned aerial vehicle flight data according to claim 5, characterized in that: The following steps are also included: S5. Generate experimental data and abnormal data to verify the reliability of the model.

Images