A short-term track deviation detection method for UAV based on four-dimensional track prediction

Abstract

The present invention discloses a method for detecting short-term track deviation of unmanned aerial vehicles based on four-dimensional track prediction, which belongs to the technical field of unmanned aerial vehicle abnormality identification, and includes the following steps: drawing a low-altitude route network, obtaining historical track data of unmanned aerial vehicles under structured routes; establishing a track data preprocessing and reconstruction model, and constructing track data containing multi-dimensional attributes such as position, speed, and heading; using the multi-dimensional data features of unmanned aerial vehicles as input, and using a bidirectional long short-term memory network algorithm to construct a short-term track prediction model for unmanned aerial vehicles; converting track deviation detection into a track point density classification problem, and reconstructing a track point data set based on historical track points and predicted track points; and establishing a method for detecting unmanned aerial vehicle track deviation based on multi-dimensional local abnormal factors between reconstructed track point data sets. The present invention is committed to eliminating the safety hazards caused by track deviation of unmanned aerial vehicles during the operation of low-altitude structured route networks, and realizing dynamic monitoring of track deviation status within a short time range.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) abnormality recognition, and in particular relates to a UAV short-term track deviation detection method based on four-dimensional track prediction. Background Art

[0002] With the gradual opening of low-altitude airspace and the rapid development of urban transportation and Internet e-commerce, rotorcraft drones have become an important means of transport in urban air transportation. They are increasingly widely used in many fields such as logistics transportation and distribution, emergency rescue, and commuting services. At the same time, the rapid development of drone positioning technology, communication technology, and big data technology has led to an exponential growth in the amount of drone track data during the flight of drones. Drone track data is a typical time series data, which can be regarded as a mapping from time to space, and objectively records the location information and time information of drone activities. In recent years, track data mining has been a research hotspot, including track clustering, track data compression, and track prediction. However, as drones are widely used in daily life, they bring convenience while also bringing many dangers. In the face of a series of dangerous events related to drones, mining the relationship between track data and drone movement based on drone track data information is an important means to achieve early warning of drone abnormal behavior. This paper defines drone track deviation as a type of drone abnormal behavior and conducts research on it. Existing track deviation detection methods can be divided into methods based on track segment similarity detection, methods based on extracting global features of trajectories, methods based on classifiers, and methods based on neural networks.

[0003] In summary, the existing track deviation detection is mostly based on historical track data, without combining the future track data of the UAV. Such methods are difficult to achieve real-time detection of track deviation. At the same time, the current detection of abnormal tracks is usually based on track clustering, and track deviation detection is achieved by extracting different types of track-related information. However, this method is difficult to identify multi-dimensional UAV track data characteristics. Summary of the invention

[0004] The purpose of the present invention is to provide a method for detecting short-term track deviation of an unmanned aerial vehicle based on four-dimensional track prediction, so as to solve the problems that the existing track deviation detection in the above-mentioned technology is difficult to realize real-time detection of track deviation and difficult to identify multi-dimensional unmanned aerial vehicle track data characteristics.

[0005] To achieve the above object, the present invention provides a method for detecting short-term track deviation of an unmanned aerial vehicle based on four-dimensional track prediction, comprising the following steps:

[0006] Step 1: Draw a low-altitude route network and obtain the historical track data of UAV operation under the structured route;

[0007] Step 2: Establish a track data preprocessing and reconstruction model to construct track data containing multi-dimensional attributes such as position, speed, and heading;

[0008] Step 3: Taking the multi-dimensional data features of the UAV as input, a bidirectional long short-term memory network algorithm is used to build a UAV short-term track prediction model, and at the same time, the maximum time span of the track prediction model with acceptable prediction accuracy is calculated;

[0009] Step 4: Reconstruct the track data set based on the historical track points and the predicted track points, and transform the track deviation detection into a track density classification problem;

[0010] Step 5: Calculate the multi-dimensional local anomaly factors between the reconstructed track point data sets, establish a UAV track deviation detection method, and realize dynamic monitoring of the track deviation status within a short time range.

[0011] Preferably, the specific process of step 1 is as follows: draw a low-altitude route network, determine the UAV flight area, collect the historical track data of the UAV running under the structured route, and assume that T is the historical track set of all UAVs, which contains N historical tracks, expressed as

[0012] T={T 1 ,T 2 ,...,T k ,...,T N}

[0013] Where, T k is the kth track in T;

[0014] Each track contains n track points

[0015] T k ={S k1 ,S k2 ,...,S ki ,...,S kn}

[0016] In the formula, S ki It is T k The i-th track point in ;

[0017] The features contained in each track point are expressed as

[0018]

[0019] In the formula, Respectively represent the longitude (°), latitude (°), altitude (m), v ki ,d ki Indicates the speed (m / s) and heading (rad) of the drone’s track point.

[0020] Preferably, the specific process of step 2 is as follows:

[0021] Step 2.1, delete the tracks with fewer track points, and for the tracks with fewer missing track points, use the Bezier curve cubic spline interpolation method to divide the tracks into five components of longitude, latitude, altitude, speed and heading related to time t, and solve them by interpolation method;

[0022] Step 2.2: The track data is preprocessed by using the maximum and minimum normalization method before inputting into the model, so that the range of eigenvalues ​​in the model input vector is consistent;

[0023] Step 2.3: In each prediction, the input window only predicts the value of one future track point, and the UAV track is reconstructed by single-step prediction.

[0024] Preferably, the specific process of step 3 is as follows:

[0025] Step 3.1, based on the preprocessed UAV track data, taking the multi-dimensional data features of the UAV as input, establish a unidirectional long short-term memory network algorithm model;

[0026] Step 3.2: Combine the forward network to forward calculate the hidden vector And the backward network reversely calculates the hidden vector Establish a bidirectional long short-term memory network algorithm track prediction model to capture the data characteristics between the two directions

[0027]

[0028] Step 3.3: Calculate the attenuation value of the track prediction model in the future and obtain the maximum time span T under the prediction accuracy. max .

[0029] Preferably, the specific process of step 4 is as follows:

[0030] Step 4.1: Use the position, speed and heading features in the detected track data to predict the track point at time T+1 The new data set is formed with the track points at the historical time T+1

[0031] Step 4.2: Calculate the track points predicted at time T+1 in the data set and the track point S at the historical time T+1 k(t+1) The Euclidean distance between S k(t+1) The coordinates of and

[0032]

[0033] Step 4.3: Input the data set into step 5 for detection. After each detection, increase the time by 1, and then reconstruct a new track data set for detection until T max The moment is over.

[0034] Preferably, the specific process of step 5 is as follows:

[0035] Step 5.1, determine the number of neighbors and threshold of the multi-dimensional local anomaly factor algorithm in the short-term track deviation detection model;

[0036] Step 5.2: Based on the Euclidean distance between each track point in the reconstructed data set, the reconstructed data set is tested and the multi-dimensional local anomaly factor value of each track point is calculated;

[0037] Step 5.3: Output the results of anomaly detection of each track point, compare the multi-dimensional local anomaly factor value and threshold of the track point, and the track point exceeding the threshold is regarded as an abnormal track point, and the track of the abnormal point is judged as deviating from the track.

[0038] Therefore, the present invention adopts the above-mentioned UAV short-term track deviation detection method based on four-dimensional track prediction, which has the following beneficial effects: track deviation detection is performed based on track point density, and the multi-dimensional characteristics of a single abnormal point in the track are effectively identified, ultimately meeting the needs of fast and accurate UAV abnormal track identification, and providing new ideas for ensuring UAV flight and promoting the innovative application and large-scale development of unmanned aviation.

[0039] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart for the overall implementation of the present invention;

[0041] Figure 2 This is a schematic diagram of UAV flight in structured airspace of the present invention;

[0042] Figure 3 Schematic diagram of the Bi-LSTM neural network of the present invention;

[0043] Figure 4 It is a schematic diagram of LOF algorithm;

[0044] Figure 5 The present invention is a flowchart of the method. DETAILED DESCRIPTION

[0045] Example

[0046] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1-5 , a method for detecting short-term track deviation of a UAV based on four-dimensional track prediction, comprising the following steps:

[0048] Step 1: Draw a low-altitude route network, determine the UAV flight area, and collect historical track data of the UAV operating under the structured route. The specific process is as follows:

[0049] Draw a low-altitude route network, determine the UAV flight area, and collect the historical track data of the UAV operating under the structured route. Assume that T is the historical track set of all UAVs, which contains N historical tracks, expressed as:

[0050] T={T 1 ,T 2 ,...,T k ,...,T N}

[0051] Where, T k is the kth track in T;

[0052] Each track contains n track points:

[0053] T k ={S k1 ,S k2 ,...,S ki ,...,S kn}

[0054] In the formula, S ki It is T k The i-th track point in ;

[0055] The features contained in each track point are expressed as:

[0056]

[0057] In the formula, Respectively represent the longitude (°), latitude (°), altitude (m), v ki ,d ki Indicates the speed (m / s) and heading (rad) of the drone’s track point.

[0058] Step 2: Establish a track data preprocessing and reconstruction model to construct track data containing multi-dimensional attributes such as position, speed, and heading. The specific process is as follows:

[0059] Step 2.1, delete the tracks with fewer track points. For the tracks with relatively few missing track points, use the B-spline cubic spline interpolation method, where the track is divided into five components of longitude, latitude, altitude, speed and heading related to time t, and the interpolation method is used to solve it. The B-spline curve is defined as:

[0060]

[0061] Among them, P i are the vertices of the characteristic polygon, and B is called the basis function;

[0062] Step 2.2: Before inputting the model, the track data is preprocessed using the max-min normalization method to make the eigenvalue range in the model input vector consistent;

[0063]

[0064] Among them, ξ is the multi-dimensional trajectory parameter vector T of the UAV kt The elements in mainly include longitude, latitude, altitude, speed and heading; min is the minimum value in the track data set, ξ max is the maximum value in the track data set; ξ' is the standardized UAV track data.

[0065] Similarly, the output of the predicted drone track position needs to be denormalized and decoded:

[0066]

[0067] Step 2.3: Reconstruct the UAV track by predicting only one track point value in the future in the input window at each prediction, that is, by single-step prediction. The specific process is as follows:

[0068] (1) Set the input of each prediction to be the UAV trajectory parameter vector of two adjacent timestamps, and the prediction result to be the UAV trajectory parameter vector of the next timestamp:

[0069]

[0070] In the formula, represents the UAV four-dimensional trajectory prediction model, T t is the predicted output UAV moment trajectory parameter vector, T t-1 ,T t-2 Represents the two adjacent timestamps of the UAV track parameter vector input into the prediction model;

[0071] (2) Based on the predicted data, the trajectory parameters of the UAV after m time stamps in the future are predicted recursively:

[0072]

[0073] Among them, each time the prediction model is called, m is the prediction time span, that is, the time difference between the predicted track point and the current position. Its recursive meaning is: through the known UAV track parameter vector T t-1 ,T t Predict the next trajectory parameter vector T t+1 , then according to T t ,T t+1 Prediction T t+2 , and so on until T t+m .

[0074] Step 3: Using the multi-dimensional data features of the UAV as input, a bidirectional long short-term memory network algorithm is used to build a UAV short-term track prediction model, and at the same time calculate the maximum time span of the track prediction model under acceptable prediction accuracy. The specific process is as follows:

[0075] Step 3.1: Based on the preprocessed UAV track data, a unidirectional LSTM model is established with the multi-dimensional data features of the UAV as input;

[0076] The input data structure of track prediction is:

[0077]

[0078] View h through the forget gate t and x t To determine the cell state C t , decide whether to keep the information:

[0079]

[0080] The output gate determines the updated value and cell state inside the cell:

[0081]

[0082]

[0083] By multiplying the cell state of the previous layer by the forgetting vector point by point, we get the updated cell state:

[0084]

[0085] Finally, the output gate determines the output result.

[0086]

[0087]

[0088] It includes the long-term state, short-term state and output of the unit in a single instance at each time iteration, x t Input for the current time, h t is the current output, C t Represents the cell state, tanh is the hyperbolic tangent activation function, and σ is the Sigmoid activation function.

[0089] Step 3.2: Combine the forward network to forward calculate the hidden vector The backward network reversely calculates the hidden vector A Bi-LSTM track prediction model is established to capture the data features between the two directions.

[0090]

[0091] The output of the track prediction is the predicted position after a fixed period of time, including the longitude, latitude and altitude of the predicted position point. Therefore, the output data structure of the track prediction is:

[0092]

[0093] Step 3.3: Calculate the loss value of the track prediction model in the future and obtain the maximum time span T with acceptable prediction accuracy. max .

[0094] Step 4: Reconstruct the track data set based on historical track points and predicted track points, and transform the track deviation detection into a track density classification problem. The specific process is as follows:

[0095] Step 4.1: Use the position, speed and heading features in the detected track data to predict the track point at time T+1 The new data set is formed with the track points at the historical time T+1

[0096] Step 4.2: Calculate the track points predicted at time T+1 in the data set The track point S at the historical time T+1 k(t+1) The Euclidean distance between S k(t+1) The coordinates of and

[0097]

[0098] Step 4.3: Input the data set into step 5 for detection. After each detection, increase the time by 1, and then reconstruct a new track data set for detection until T max The moment is over.

[0099] Step 5: Calculate the multi-dimensional local anomaly factors between the reconstructed track point data sets, establish a UAV track deviation detection method, and realize dynamic monitoring of the track deviation state within a short time range. The specific process is as follows:

[0100] Step 5.1: Determine the number of neighbors k and the threshold ε of the LOF algorithm in the short-term track deviation detection model.

[0101] Step 5.2: Based on the Euclidean distance between each track point in the reconstructed data set, the reconstructed data set is tested and the LOF value of each track point is calculated.

[0102] (1) Calculate the predicted track points The value of k:

[0103]

[0104] in, Centralize and predict track points for historical track data The closest track point to the kth track point;

[0105] (2) Calculate the predicted track points The nearest track points of :

[0106]

[0107] (3) Calculate the predicted track points To historical track point S t The reachable distance:

[0108]

[0109] (4) Calculate the predicted track points Neighbor density:

[0110]

[0111] in: replace Smooth the neighbor density. The larger the k value, the more obvious the smoothing effect.

[0112] (5) Calculate the predicted track points Local anomaly factor:

[0113]

[0114] In the formula, the track point With track point St The greater the ratio of the neighbor density of the neighboring tracks, the The greater the abnormality

[0115] Step 5.3: Output the results of anomaly detection for each track point, compare the LOF value of the track point with the threshold ε, and the track points exceeding ε are considered abnormal track points, and the track containing abnormal points is judged as deviating from the track.

[0116] Therefore, the present invention adopts the above-mentioned UAV short-term track deviation detection method based on four-dimensional track prediction. Based on the four-dimensional track prediction results of the UAV, the historical track points and predicted track point data sets of the UAV are reconstructed, and the historical track data information and future track data of the UAV are integrated to construct a track point density anomaly detection model, thereby realizing short-term dynamic monitoring of the UAV and eliminating the safety hazards caused by track deviation of the UAV during the operation of the low-altitude structured route network.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for detecting short-term track deviation of UAV based on four-dimensional track prediction. It is characterized in that The following steps are involved: Step 1: Draw a low-altitude route network and obtain the historical track data of UAV operation under the structured route; Step 2: Establish a track data preprocessing and reconstruction model to construct track data containing multi-dimensional attributes such as position, speed, and heading; Step 3: Taking the multi-dimensional data features of the UAV as input, a bidirectional long short-term memory network algorithm is used to build a UAV short-term track prediction model, and at the same time, the maximum time span of the track prediction model with acceptable prediction accuracy is calculated; Step 4: Reconstruct the track data set based on historical track points and predicted track points, and transform the track deviation detection into a track density classification problem; Step 5: Calculate the multi-dimensional local anomaly factors between the reconstructed track point data sets, establish a UAV track deviation detection method, and realize dynamic monitoring of the track deviation state within a short time range; The specific process of step 4 is as follows: Step 4.1: Use the position, speed and heading features in the detected track data to predict the track point at time T+1 The new data set is formed with the track points at the historical time T+1 Step 4.2: Calculate the track points predicted at time T+1 in the data set and the track point S at the historical time T+1 k(t+1) The Euclidean distance between S k(t+1) The coordinates of and Step 4.3: Input the data set into step 5 for detection. After each detection, increase the time by 1, and then reconstruct a new track data set for detection until T max The moment is over; The specific process of step 5 is as follows: Step 5.1, determine the number of neighbors and threshold of the multi-dimensional local anomaly factor algorithm in the short-term track deviation detection model; Step 5.2: Based on the Euclidean distance between each track point in the reconstructed data set, the reconstructed data set is tested to calculate the multi-dimensional local anomaly factor value of each track point; Step 5.3: Output the results of anomaly detection of each track point, compare the multi-dimensional local anomaly factor value and threshold of the track point, and the track point exceeding the threshold is regarded as an abnormal track point, and the track of the abnormal point is judged as deviating from the track.

2. According to claim 1, a method for detecting short-term track deviation of an unmanned aerial vehicle based on four-dimensional track prediction, It is characterized in that The specific process of step 1 is as follows: draw a low-altitude route network, determine the UAV flight area, and collect the historical track data of the UAV operating under the structured route. Assume that T is the historical track set of all UAVs, which contains N historical tracks, expressed as T={T 1 ,T 2 ,...,T k ,...,T N } Where, T k is the kth track in T; Each track contains n track points T k ={S k1 ,S k2 ,...,S ki ,...,S kn } In the formula, S ki It is T k The i-th track point in ; The features contained in each track point are expressed as In the formula, λ ki ,h ki Respectively represent the longitude of the drone track point, unit: °, latitude, unit: °, altitude, unit: m, v ki ,d ki Indicates the speed of the drone track point in m / s and the heading in rad.

3. According to claim 1, a method for detecting short-term track deviation of an unmanned aerial vehicle based on four-dimensional track prediction, It is characterized in that The specific process of step 2 is as follows: Step 2.1, delete the tracks with fewer track points, and for the tracks with fewer missing track points, use the Bezier curve cubic spline interpolation method to divide the tracks into five components of longitude, latitude, altitude, speed and heading related to time t, and solve them by interpolation method; Step 2.2: The track data is preprocessed by using the maximum and minimum normalization method before inputting into the model, so that the range of eigenvalues ​​in the model input vector is consistent; Step 2.3: In each prediction, the input window only predicts the value of one future track point, and the UAV track is reconstructed by single-step prediction.

4. According to claim 1, a method for detecting short-term track deviation of an unmanned aerial vehicle based on four-dimensional track prediction, It is characterized in that The specific process of step 3 is as follows: Step 3.1, based on the preprocessed UAV track data, taking the multi-dimensional data features of the UAV as input, establish a unidirectional long short-term memory network algorithm model; Step 3.2: Combine the forward network to forward calculate the hidden vector And the backward network reversely calculates the hidden vector Establish a bidirectional long short-term memory network algorithm track prediction model to capture the data characteristics between the two directions Step 3.3: Calculate the attenuation value of the track prediction model in the future and obtain the maximum time span T under the prediction accuracy. max .

Images