Aviation hazard wind field detection method and device based on nonlinear residual network

Abstract

The present invention discloses a wind field detection method, specifically relates to an aviation hazardous wind field detection method and device based on a nonlinear residual network. The method includes: using a radar to detect an aircraft take-off and landing channel, obtaining Doppler velocity sampling data of the aircraft take-off and landing channel, and recording the hazardous moment of encountering a hazardous wind field; establishing a hazard feature library based on the second-order structural function of the Doppler velocity sampling data and the traditional hazard assessment features corresponding to the Doppler velocity sampling data; extracting representative physical features from the hazard feature library; training a nonlinear residual network based on hazardous wind field data, non-hazardous wind field data, and representative physical features corresponding to hazardous wind field data and non-hazardous wind field data; performing aviation hazardous wind field detection based on the nonlinear residual network to obtain detection results. The embodiment of the present invention overcomes the problems of interpretability and insufficient label data faced by traditional neural network methods, and has a high accuracy rate for hazardous wind field detection.

Description

Technical Field

[0001] The present invention relates to a wind field detection method, and in particular to an aviation hazard wind field detection method and device based on a nonlinear residual network. Background Art

[0002] Wind shear and turbulence, which may be present in low-altitude wind fields, pose a serious threat to aviation safety. The variations in wind speed and direction induced by different types of airflow have significantly different characteristics. For example, turbulence can cause rapid fluctuations in wind speed and direction within a meter, while wind shear can cause variations in wind speed and direction that persist for more than several nautical miles.

[0003] Current wind field hazard assessment methods focus on different types of hazardous wind fields. For turbulence, the most commonly used hazard assessment features include average wind speed, turbulent dissipation rate, and velocity fluctuation. For wind shear, the most commonly used hazard assessment features include the existing F-factor and S-factor. These hazard features are only applicable to specific types of hazardous airflow, and the corresponding alarm thresholds are not adaptable to airports with different geographical locations and meteorological conditions. Consequently, the accuracy of hazardous wind field detection is insufficient in practical applications.

[0004] In recent years, deep learning methods have made some progress in areas such as image classification, speech recognition, and change detection. However, neural networks face the following two major challenges: 1) Lack of interpretability: In the task of hazardous wind field detection, it is impossible to explain which physical features are helpful for hazardous wind field detection; 2) Insufficient labeled data: In the task of hazardous wind field detection, due to insufficient labeled data, the network cannot provide accurate hazardous wind field detection results through training. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the above-mentioned methods commonly used in the prior art all have the problems of insufficient interpretability and insufficient label data. In order to solve the above problems, the present invention provides an aviation hazard wind field detection method and device based on a nonlinear residual network.

[0006] The present invention includes:

[0007] In a first aspect, an embodiment of the present invention provides an aviation hazard wind field detection method based on a nonlinear residual network, comprising:

[0008] Using radar to detect aircraft take-off and landing corridors, obtaining Doppler velocity sampling data of the aircraft take-off and landing corridors, and recording the hazardous time when encountering hazardous wind fields;

[0009] Establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor;

[0010] extracting representative physical features from the hazard feature library;

[0011] A nonlinear residual network is obtained by training based on hazardous wind field data, non-hazardous wind field data, representative physical features corresponding to the hazardous wind field data, and representative physical features corresponding to the non-hazardous wind field data, wherein the hazardous wind field data is data corresponding to the hazardous moment in the Doppler velocity sampling data, and the non-hazardous wind field data is data not corresponding to the hazardous moment in the Doppler velocity sampling data;

[0012] Aviation hazard wind field detection is performed based on the nonlinear residual network to obtain detection results.

[0013] Optionally, performing aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result includes:

[0014] Use radar to detect aircraft take-off and landing channels and obtain target Doppler velocity sampling data;

[0015] Calculating corresponding representative physical characteristics based on the target Doppler velocity sampling data;

[0016] The target Doppler velocity sampling data and representative physical features corresponding to the target Doppler velocity sampling data are input into the nonlinear residual network to perform hazardous wind field detection to obtain a detection result.

[0017] Optionally, before training a nonlinear residual network based on the hazardous wind farm data, the non-hazardous wind farm data, the representative physical features corresponding to the hazardous wind farm data, and the representative physical features corresponding to the non-hazardous wind farm data, the method further includes:

[0018] Marking the first data as the hazardous wind farm data, and marking at least part of the second data as non-hazardous wind farm data;

[0019] The first data is the data corresponding to the hazardous moment in the Doppler velocity sampling data, the second data is the data in the Doppler velocity sampling data other than the first data, and the number of the hazardous wind field data is the same as the number of the non-hazardous wind field data.

[0020] Optionally, before establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, the method further includes:

[0021] A second-order structure function of the Doppler velocity sampling data is calculated based on the Doppler velocity sampling data, and a traditional hazard assessment feature corresponding to the Doppler velocity sampling data is calculated based on the Doppler velocity sampling data.

[0022] Optionally, the second-order structure function of the Doppler velocity sampling data satisfies:

[0023] s(r)=mean([v(x)-v(x+r)] 2 );

[0024] Among them, v(x) is used to represent the Doppler velocity sampling data detected by the radar at the position x of the aircraft take-off and landing channel, v(x+r) is used to represent the Doppler velocity sampling data detected by the radar at the position x+r of the aircraft take-off and landing channel, r represents the distance between x and x+r, and mean() represents the calculated average value.

[0025] The Doppler velocity average value satisfies:

[0026] M = mean(|v|);

[0027] Wherein, v represents a set of Doppler velocity sampling data;

[0028] The turbulent energy dissipation rate satisfies:

[0029]

[0030] Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data;

[0031] The speed fluctuation satisfies:

[0032] R = max(v) - min(v);

[0033] Wherein, max(v) is the maximum value of the Doppler velocity sampling data, and min(v) is the minimum value of the Doppler velocity sampling data;

[0034] The F-factor satisfies:

[0035]

[0036] Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed along the aircraft take-off and landing channel, V app is the aircraft airspeed, g is the acceleration due to gravity;

[0037] The S-factor satisfies:

[0038]

[0039] Where, ΔV i is the increment or decrement of the Doppler velocity sampling data, R i is the length of the corresponding slope in the aircraft take-off and landing channel, and the subscript i represents the i-th slope.

[0040] Optionally, the nonlinear residual network satisfies:

[0041]

[0042] Wherein, v is used to characterize the input Doppler velocity sampling data, z is used to characterize the representative physical characteristics of the input, ReLU() represents the rectified linear unit activation function, sigmoid[] represents the logarithmic probability function, {W1, W2, b1, b2} are the parameters of the nonlinear residual network, Indicates the output detection result.

[0043] Optionally, the loss function for training the nonlinear residual network satisfies:

[0044]

[0045] is the mean square error of the detection result, y i is the true label of the Doppler velocity sampling data of the i-th group, is the regularization term, and λ is the regularization coefficient.

[0046] In a second aspect, an embodiment of the present invention provides an aviation hazard wind field detection device based on a nonlinear residual network, comprising:

[0047] A detection module is used to detect aircraft take-off and landing channels using radar, obtain Doppler velocity sampling data of the aircraft take-off and landing channels, and record the time when the aircraft encounters a hazardous wind field;

[0048] An establishment module is used to establish a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor;

[0049] An extraction module, configured to extract representative physical features from the hazard feature library;

[0050] a training module, configured to train a nonlinear residual network based on hazardous wind field data, non-hazardous wind field data, representative physical features corresponding to the hazardous wind field data, and representative physical features corresponding to the non-hazardous wind field data, wherein the hazardous wind field data is data corresponding to the hazardous moment in the Doppler velocity sampling data, and the non-hazardous wind field data is data not corresponding to the hazardous moment in the Doppler velocity sampling data;

[0051] The detection module is used to perform aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result.

[0052] In a third aspect, an embodiment of the present invention provides an electronic device comprising: a memory, a processor, and a program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps of the aviation hazard wind field detection method based on a nonlinear residual network as described in the first aspect.

[0053] In a fourth aspect, an embodiment of the present invention provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps of the aviation hazard wind field detection method based on a nonlinear residual network as described in the first aspect.

[0054] The beneficial effects of the present invention are as follows: in an embodiment of the present invention, radar is used to detect aircraft take-off and landing channels, Doppler velocity sampling data of the aircraft take-off and landing channels is obtained, and the time of encountering hazardous wind fields is recorded; a hazard feature library is established based on the second-order structure function of the Doppler velocity sampling data and the traditional hazard assessment features corresponding to the Doppler velocity sampling data; representative physical features are extracted from the hazard feature library; a nonlinear residual network is trained based on hazardous wind field data, non-hazardous wind field data, representative physical features corresponding to the hazardous wind field data, and representative physical features corresponding to the non-hazardous wind field data; aviation hazardous wind field detection is performed based on the nonlinear residual network to obtain detection results. The nonlinear residual network constructed by the present invention has a small number of layers and is simple to implement. In addition, the number of network parameters is further reduced by a dimensionality reduction method; the nonlinear residual network combines linear feature selection and nonlinear residual terms, and can explain which physical features are helpful for hazardous wind field detection through linear feature selection. It has the advantages of few network parameters and strong interpretability, and has a high accuracy rate in hazardous wind field detection in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Attachment Figure 1 One of the flow charts of the aviation hazard wind field detection method based on nonlinear residual network provided in an embodiment of the present invention;

[0056] Attachment Figure 2 Flowchart 2 of the aviation hazard wind field detection method based on nonlinear residual network provided in an embodiment of the present invention;

[0057] Attachment Figure 3 A schematic diagram of radar scanning provided by an embodiment of the present invention;

[0058] Attachment Figure 4 A schematic diagram of Doppler velocity sampling data provided by an embodiment of the present invention;

[0059] Attachment Figure 5 A schematic diagram of constructing a harmful feature library provided by an embodiment of the present invention;

[0060] Attachment Figure 6 A schematic diagram of a nonlinear residual network structure provided by an embodiment of the present invention;

[0061] Attachment Figure 7a A box plot of key success indices for detecting hazardous wind farms using different methods provided by embodiments of the present invention;

[0062] Attachment Figure 7b A distribution diagram of the median of the key success index for detecting hazardous wind farms using different methods provided in embodiments of the present invention;

[0063] Attachment Figure 8 A schematic diagram of an aviation hazard wind field detection device based on a nonlinear residual network provided in an embodiment of the present invention;

[0064] Attachment Figure 9 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In the embodiments of the present application, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In the embodiments of the present application, the term "multiple" refers to two or more, and other quantifiers are similar. The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0066] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0067] See Figure 1 The present invention proposes a method for detecting aviation hazardous wind fields based on a nonlinear residual network. A hazard feature library is constructed based on traditional hazard assessment features. The establishment of a nonlinear residual network solves the problems of insufficient interpretability of neural network detection results and insufficient label data, which can greatly improve the robustness of hazardous wind field assessment in practical applications. Figure 1 A schematic flow chart of a method for detecting aviation hazard wind fields based on a nonlinear residual network provided in an embodiment of the present invention.

[0068] like Figure 1 As shown, this method specifically includes the following steps:

[0069] Step 101: Use radar to detect the aircraft take-off and landing channel, obtain Doppler velocity sampling data of the aircraft take-off and landing channel, and record the hazardous time when encountering a hazardous wind field.

[0070] Step 102: establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average value, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor.

[0071] Step 103: extract representative physical features from the hazard feature library.

[0072] Step 104: A nonlinear residual network is trained based on the hazardous wind field data, the non-hazardous wind field data, the representative physical features corresponding to the hazardous wind field data, and the representative physical features corresponding to the non-hazardous wind field data. The hazardous wind field data is the data in the Doppler velocity sampling data corresponding to the hazardous moment, and the non-hazardous wind field data is the data in the Doppler velocity sampling data not corresponding to the hazardous moment.

[0073] Step 105 : Perform aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result.

[0074] Optionally, in some embodiments, step 105 includes:

[0075] Use radar to detect aircraft take-off and landing channels and obtain target Doppler velocity sampling data;

[0076] Calculating corresponding representative physical characteristics based on the target Doppler velocity sampling data;

[0077] The target Doppler velocity sampling data and representative physical features corresponding to the target Doppler velocity sampling data are input into the nonlinear residual network to perform hazardous wind field detection to obtain a detection result.

[0078] By using radar to detect aircraft takeoff and landing corridors, we can obtain radar observation data at different locations along the corridor, namely Doppler velocity sampling data. A set of Doppler velocity sampling data v is used to represent the Doppler velocity detected by the radar at different locations along the corridor.

[0079] For the convenience of description, in the embodiment of this application, the constructed hazard signature library is recorded as:

[0080] x=[s(r1),s(r2),…s(r N ),M,ε,R,F,S];

[0081] where [s(r1),s(r2),…s(r N )] represents the distances corresponding to different distances (denoted as r1, r2, ... r N), M represents the average value of Doppler velocity sampling data, ε represents the turbulent energy dissipation rate, R represents the velocity fluctuation, F and S are F-factor and S-factor respectively.

[0082] Optionally, in some embodiments, before step 102, the method further includes:

[0083] A second-order structure function of the Doppler velocity sampling data is calculated based on the Doppler velocity sampling data, and a traditional hazard assessment feature corresponding to the Doppler velocity sampling data is calculated based on the Doppler velocity sampling data.

[0084] Optionally, the second-order structure function of the Doppler velocity sampling data satisfies:

[0085] s(r)=mean([v(x)-v(x+r)] 2 );

[0086] Among them, v(x) is used to represent the Doppler velocity sampling data detected by the radar at the position x of the aircraft take-off and landing channel, v(x+r) is used to represent the Doppler velocity sampling data detected by the radar at the position x+r of the aircraft take-off and landing channel, r represents the distance between x and x+r, and mean() represents the calculated average value.

[0087] Optionally, the Doppler velocity average value satisfies:

[0088] M = mean(|v|);

[0089] Wherein, v represents a set of Doppler velocity sampling data;

[0090] Optionally, the turbulent energy dissipation rate satisfies:

[0091]

[0092] Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data;

[0093] Optionally, the speed fluctuation satisfies:

[0094] R = max(v) - min(v);

[0095] Wherein, max(v) is the maximum value of the Doppler velocity sampling data, and min(v) is the minimum value of the Doppler velocity sampling data;

[0096] Optionally, the F-factor satisfies:

[0097]

[0098] Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed along the aircraft take-off and landing channel, V app is the aircraft airspeed, g is the acceleration due to gravity, in this embodiment, the vertical wind speed term is ignored, and it is assumed that u x =v;

[0099] Optionally, the S-factor satisfies:

[0100]

[0101] Where, ΔV i is the increment or decrement of the Doppler velocity sampling data, R i is the length of the corresponding slope in the aircraft take-off and landing channel, and the subscript i represents the i-th slope.

[0102] After constructing the hazard feature library, a dimensionality reduction method can be used to select multiple principal components from the hazard feature library as representative physical features for hazardous wind field detection. The specific dimensionality reduction method is not limited here. For example, representative physical features can be extracted from the hazard feature library using principal component analysis. This process can be described as follows:

[0103] z = PCA(x);

[0104] Here, x and z represent the hazard signature library and representative physical characteristics, respectively, and PCA() represents principal component analysis. The specific steps are: first, decentralize and normalize x; then, calculate the eigenvalues ​​and eigenvectors of the covariance matrix of the centered x and arrange them in descending order; construct a projection matrix from the eigenvectors corresponding to the first M eigenvalues; and finally, transform x into the new space formed by the projection matrix to achieve data dimensionality reduction, thereby reducing the complexity of the present invention and improving interpretability.

[0105] Optionally, in some embodiments, the nonlinear residual network satisfies:

[0106]

[0107] Wherein, v is used to characterize the input Doppler velocity sampling data, z is used to characterize the representative physical characteristics of the input, ReLU() represents the rectified linear unit activation function, sigmoid[] represents the logarithmic probability function, {W1, W2, b1, b2} are the parameters of the nonlinear residual network, The first two terms “zW1+b1” in the above equation represent linear feature selection, and the last term “ReLU(vW2+b2)” represents the nonlinear residual term.

[0108] Optionally, in some embodiments, the loss function for training the nonlinear residual network satisfies:

[0109]

[0110] in, is the mean square error of the detection result, y i is the true label of the Doppler velocity sampling data of the i-th group (if the Doppler velocity sampling data is the marked hazardous wind field data, then y i =1; otherwise, if the Doppler velocity sampling data is the marked non-hazardous wind field data, then y i =0), is the regularization term, and λ is the regularization coefficient.

[0111] Optionally, in some embodiments, before step 104, the method further includes:

[0112] Marking the first data as the hazardous wind farm data, and marking at least part of the second data as non-hazardous wind farm data;

[0113] The first data is the data corresponding to the hazardous moment in the Doppler velocity sampling data, the second data is the data in the Doppler velocity sampling data other than the first data, and the number of the hazardous wind field data is the same as the number of the non-hazardous wind field data.

[0114] The method provided by the embodiment of the present invention obtains Doppler velocity sampling data and the hazardous moment when the pilot reports encountering a hazardous wind field, and labels the hazardous wind field data and non-hazardous wind field data accordingly. Then, a hazard feature library is established by combining traditional hazard assessment features and the second-order structure function of the Doppler velocity sampling data. A dimensionality reduction method is then used to select multiple principal components from the hazard feature library as representative physical features for hazardous wind field detection. The constructed nonlinear residual network is trained in combination with the Doppler velocity sampling data and the representative physical features to obtain a nonlinear residual network, and then the nonlinear residual network is used to complete aviation hazardous wind field detection.

[0115] Compared with the existing technology, this method uses a nonlinear residual network to detect whether Doppler velocity sampling data is hazardous from the Doppler velocity sampling data and the hazard feature library. The constructed nonlinear residual network has few layers and is simple to implement. In addition, the number of network parameters is further reduced by the dimensionality reduction method. The nonlinear residual network combines linear feature selection and nonlinear residual terms. It can explain which physical characteristics are helpful for hazardous wind field detection through linear feature selection. It has the advantages of few network parameters and strong interpretability, overcoming the difficulties of interpretability and insufficient label data faced by traditional neural network methods. Compared with traditional hazard assessment methods, the above scheme has a higher accuracy rate in hazardous wind field detection in practical applications.

[0116] See Figure 2 , the following is an example of a specific embodiment. Figure 2 As shown in FIG, the aviation hazard wind field detection method based on the nonlinear residual network specifically includes the following steps:

[0117] Step 1: Use radar to detect the aircraft take-off and landing corridor area, obtain Doppler velocity sampling data on the aircraft take-off and landing corridor, and record the time when the pilot reports encountering a hazardous wind field;

[0118] Step 2: The data corresponding to the hazardous time in the overall radar observation data are used as the marked hazardous wind field data; Doppler velocity sampling data with the same number as the marked hazardous wind field data are randomly selected from the data not corresponding to the hazardous time in the overall radar observation data as the marked non-hazardous wind field data;

[0119] Step 3: Integrate the second-order structure function of Doppler velocity sampling data and traditional hazard assessment features to establish a hazard feature library;

[0120] Step 4: Use the dimensionality reduction method to select multiple principal components from the hazard feature library as representative physical features for hazard wind field detection;

[0121] Step 5: Build a nonlinear residual network. Using steps 3 and 4, calculate the representative physical features corresponding to the labeled hazardous and non-hazardous wind field data. Combined with the Doppler velocity sampling data and the representative physical features, the constructed nonlinear residual network is trained. In practice, real-time Doppler velocity sampling data and its corresponding representative physical features are fed into the trained nonlinear residual network to detect hazardous wind fields for aviation.

[0122] The following is an example of a specific embodiment. Figure 3As shown, one or more microwave radars or lidars can be set up next to the airport runway to collect online radar observation data. The radar performs a low-elevation-angle PPI (plane position indicator) scan of the wind field area above the airport runway. The elevation angle of the radar beam can be as consistent as possible with the angle between the aircraft take-off and landing channel and the horizontal ground. Figure 3 The middle sector area is the coverage area of ​​the radar scan. The Doppler velocity data obtained by the radar scanning the area above the runway is interpolated to obtain the Doppler velocity sampling data on the aircraft takeoff and landing channel. A schematic diagram of one type of Doppler velocity sampling data can be shown as follows: Figure 4 As shown, Figure 4 Contains the first set of Doppler velocity sampling data (such as Figure 4 As shown above) and the second set of Doppler velocity sampling data (as shown above) Figure 4 As shown below, each set of Doppler velocity sampling data includes radar observation values ​​within a certain distance range from the aircraft landing point on the aircraft take-off and landing channel, with a total of L sampling points. In this embodiment, L=115.

[0123] like Figure 4 As shown in the figure, based on the Doppler velocity sampling data, the second-order structure function, average velocity, turbulent energy dissipation rate, velocity fluctuation, F factor and S factor are calculated in sequence to obtain the hazard feature library. The specific calculation method is as follows:

[0124] (1) The second-order structure function is expressed as:

[0125] s(r)=mean([v(x)-v(x+r)] 2 ) (1)

[0126] Where v(x) represents the Doppler velocity detected by the radar at position x on the aircraft take-off and landing channel, v(x+r) represents the Doppler velocity detected by the radar at position x+r on the aircraft take-off and landing channel, r represents the distance between the two points, and mean() represents the calculated average value. In this embodiment, different distances [r1, r2, ... r N ]=[100,150,…5750],N=114,the second-order structure function corresponding to different distances can be obtained by formula (1)

[0127] (2) The average speed is recorded as:

[0128] M=mean(|v|)(2)

[0129] Where v represents a set of Doppler velocity sampling data, which indicates the Doppler velocity detected by the radar at different positions on the aircraft take-off and landing channel.

[0130] (3) The turbulent energy dissipation rate is expressed as:

[0131]

[0132] Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, and in this embodiment is taken as 2, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data.

[0133] (4) The velocity fluctuation is recorded as:

[0134] R=max(v)-min(v) (4)

[0135] Wherein, max(v) and min(v) represent the maximum and minimum values ​​of the Doppler velocity sampling data, respectively.

[0136] (5) The F-factor is expressed as:

[0137]

[0138] Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed, V app is the aircraft airspeed, g is the acceleration due to gravity. In this embodiment, the vertical wind speed term is ignored and u is assumed to be x =v.

[0139] (6) The S-factor is recorded as:

[0140]

[0141] Where, ΔV i is the increment or decrement of Doppler velocity sampling data, R i is the corresponding slope length, and the subscript i represents the i-th slope.

[0142] In this embodiment, representative physical features are extracted from the hazard feature library using principal component analysis. This process can be described as follows:

[0143] z=PCA(x)(7)

[0144] Among them, PCA() represents principal component analysis, and They represent the hazard feature library and representative physical features respectively, M<<(N+5). In this embodiment, M=20 and N=114.

[0145] like Figure 5 As shown, in one embodiment, the nonlinear residual network constructed above is:

[0146]

[0147] Where v and z represent the input Doppler velocity sampling data and the representative physical features extracted from the hazard feature library, respectively; ReLU() represents the rectified linear unit activation function; sigmoid[] represents the logarithmic probability function; {W1, W2, b1, b2} are the parameters to be learned for the network. Represents the detection result output by the nonlinear residual network; the cost function of the nonlinear residual network is:

[0148]

[0149] in, is the mean square error of the detection result, denoted as y i is the true label of the i-th group of Doppler velocity sampling data (if the Doppler velocity sampling data is the marked hazardous wind field data, then y i =1; otherwise, if the Doppler velocity sampling data is marked as non-hazardous wind field data, then y i =0. ), is the regularization term, λ is the regularization coefficient.

[0150] like Figure 7a and Figure 7b As shown, this embodiment compares several different methods for detecting hazardous wind farms to detect the critical success index (CSI) of hazardous wind farms, wherein: Figure 7a Box plot of the key success index for different methods, Figure 7b The figure shows the distribution of the median of the key success index of different methods. The comparison methods include the above-mentioned aviation hazard wind field detection method based on nonlinear residual network (hereinafter referred to as the present invention), the traditional hazard wind field detection method (including average wind speed, turbulence dissipation rate, speed fluctuation, F-factor, S-factor) and the single-layer binary classification neural network method. Figure 7a and Figure 7b A comparison shows that, compared with the traditional hazardous wind field detection method and the single-layer binary classification neural network method, the aviation hazardous wind field detection method based on the nonlinear residual network provided by the embodiment of the present invention can give the detection results of the hazardous wind field with the highest critical success index.

[0151] See Figure 8 The embodiment of the present application further provides an aviation hazard wind field detection device 800 based on a nonlinear residual network, comprising:

[0152] The detection module 801 is used to detect the aircraft take-off and landing channel using radar, obtain Doppler velocity sampling data of the aircraft take-off and landing channel, and record the time when the aircraft encounters the hazardous wind field;

[0153] Establishing module 802, for establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor;

[0154] Extraction module 803, used to extract representative physical features from the hazard feature library;

[0155] A training module 804 is configured to train a nonlinear residual network based on hazardous wind field data, non-hazardous wind field data, and representative physical features corresponding to the hazardous wind field data and the non-hazardous wind field data, wherein the hazardous wind field data is data in the Doppler velocity sampling data corresponding to the hazardous moment, and the non-hazardous wind field data is data in the Doppler velocity sampling data not corresponding to the hazardous moment;

[0156] The detection module 805 is used to perform aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result.

[0157] Optionally, the detection module 805 includes:

[0158] The detection unit is used to detect the aircraft take-off and landing channels using radar and obtain target Doppler velocity sampling data;

[0159] a calculation unit, configured to calculate corresponding representative physical characteristics based on the target Doppler velocity sampling data;

[0160] The input module is used to input the target Doppler velocity sampling data and the representative physical characteristics corresponding to the target Doppler velocity sampling data into the nonlinear residual network to perform hazardous wind field detection and obtain a detection result.

[0161] Optionally, the nonlinear residual network-based aviation hazard wind field detection device 800 further includes:

[0162] a marking module, configured to mark the first data as the hazardous wind farm data, and mark at least part of the second data as non-hazardous wind farm data;

[0163] The first data is the data corresponding to the hazardous moment in the Doppler velocity sampling data, the second data is the data in the Doppler velocity sampling data other than the first data, and the number of the hazardous wind field data is the same as the number of the non-hazardous wind field data.

[0164] Optionally, the nonlinear residual network-based aviation hazard wind field detection device 800 further includes:

[0165] A calculation module is used to calculate the second-order structure function of the Doppler velocity sampling data based on the Doppler velocity sampling data, and calculate the traditional hazard assessment feature corresponding to the Doppler velocity sampling data based on the Doppler velocity sampling data.

[0166] Optionally, the second-order structure function of the Doppler velocity sampling data satisfies:

[0167] s(r)=mean([v(x)-v(x+r)] 2 );

[0168] Among them, v(x) is used to represent the Doppler velocity sampling data detected by the radar at the position x of the aircraft take-off and landing channel, v(x+r) is used to represent the Doppler velocity sampling data detected by the radar at the position x+r of the aircraft take-off and landing channel, r represents the distance between x and x+r, and mean() represents the calculated average value.

[0169] The Doppler velocity average value satisfies:

[0170] M = mean(|v|);

[0171] Wherein, v represents a set of Doppler velocity sampling data;

[0172] The turbulent energy dissipation rate satisfies:

[0173]

[0174] Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data;

[0175] The speed fluctuation satisfies:

[0176] R = max(v) - min(v);

[0177] Wherein, max(v) is the maximum value of the Doppler velocity sampling data, and min(v) is the minimum value of the Doppler velocity sampling data;

[0178] The F-factor satisfies:

[0179]

[0180] Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed along the aircraft take-off and landing channel, V app is the aircraft airspeed, g is the acceleration due to gravity;

[0181] The S-factor satisfies:

[0182]

[0183] Where, ΔV i is the increment or decrement of the Doppler velocity sampling data, R i is the length of the corresponding slope in the aircraft take-off and landing channel, and the subscript i represents the i-th slope.

[0184] Optionally, the nonlinear residual network satisfies:

[0185]

[0186] Wherein, v is used to characterize the input Doppler velocity sampling data, z is used to characterize the representative physical characteristics of the input, ReLU() represents the rectified linear unit activation function, sigmoid[] represents the logarithmic probability function, {W1, W2, b1, b2} are the parameters of the nonlinear residual network, Indicates the output detection result.

[0187] Optionally, the loss function for training the nonlinear residual network satisfies:

[0188]

[0189] is the mean square error of the detection result, y i is the true label of the Doppler velocity sampling data of the i-th group, is the regularization term, and λ is the regularization coefficient.

[0190] The aviation hazard wind field detection device 800 based on a nonlinear residual network provided in the embodiment of the present application can execute the above method embodiment. Its implementation principle and technical effects are similar, and this embodiment will not be repeated here.

[0191] It should be noted that the division of units in the embodiments of the present application is schematic and is merely a logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0192] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0193] like Figure 9 As shown, an embodiment of the present application provides an electronic device 900, including: a memory 902, a processor 901, and a program stored in the memory 902 and executable on the processor 901; the processor 901 is configured to read the program in the memory 902 to implement the steps in the aforementioned method for detecting aviation hazard wind fields based on a nonlinear residual network.

[0194] An embodiment of the present application also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, the various processes of the above-mentioned embodiment of the aviation hazard wind field detection method based on a nonlinear residual network are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here. Among them, the readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (such as compact disk (CD), digital video disc (DVD), Blu-ray disc (BD), high-definition versatile disc (HVD), etc.), and semiconductor memory (such as read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read only memory (EEPROM), non-volatile memory (NAND FLASH), solid state drive (SSD)), etc.

[0195] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0196] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD-ROM), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0197] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A method for detecting aviation hazard wind fields based on nonlinear residual networks, characterized by: include: Using radar to detect aircraft take-off and landing corridors, obtaining Doppler velocity sampling data of the aircraft take-off and landing corridors, and recording the hazardous time when encountering hazardous wind fields; Establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor; extracting representative physical features from the hazard feature library; A nonlinear residual network is obtained by training based on hazardous wind field data, non-hazardous wind field data, representative physical features corresponding to the hazardous wind field data, and representative physical features corresponding to the non-hazardous wind field data, wherein the hazardous wind field data is data corresponding to the hazardous moment in the Doppler velocity sampling data, and the non-hazardous wind field data is data not corresponding to the hazardous moment in the Doppler velocity sampling data; Performing aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result; Before establishing a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, the method further includes: calculating a second-order structure function of the Doppler velocity sampling data based on the Doppler velocity sampling data, and calculating a traditional hazard assessment feature corresponding to the Doppler velocity sampling data based on the Doppler velocity sampling data; The second-order structure function of the Doppler velocity sampling data satisfies: s(r)=mean([v(x)-v(x+r)] 2 ); Wherein, v(x) is used to represent the Doppler velocity sampling data detected by the radar at the position x of the aircraft take-off and landing channel, v(x+r) is used to represent the Doppler velocity sampling data detected by the radar at the position x+r of the aircraft take-off and landing channel, r represents the distance between x and x+r, and mean() represents the calculated average value; The Doppler velocity average value satisfies: M = mean(|v|); Wherein, v represents a set of Doppler velocity sampling data; The turbulent energy dissipation rate satisfies: Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data; The speed fluctuation satisfies: R = max(v) - min(v); Wherein, max(v) is the maximum value of the Doppler velocity sampling data, and min(v) is the minimum value of the Doppler velocity sampling data; The F-factor satisfies: Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed along the aircraft take-off and landing channel, V app is the aircraft airspeed, g is the acceleration due to gravity; The S-factor satisfies: Among them, △V i is the increment or decrement of the Doppler velocity sampling data, R i is the length of the corresponding slope in the aircraft take-off and landing channel, and the subscript i represents the i-th slope.

2. The method according to claim 1, wherein: The performing of aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result includes: Use radar to detect aircraft take-off and landing channels and obtain target Doppler velocity sampling data; Calculating corresponding representative physical characteristics based on the target Doppler velocity sampling data; The target Doppler velocity sampling data and representative physical features corresponding to the target Doppler velocity sampling data are input into the nonlinear residual network to perform hazardous wind field detection to obtain a detection result.

3. The method according to claim 1, wherein: Before training a nonlinear residual network based on the hazardous wind farm data, the non-hazardous wind farm data, the representative physical features corresponding to the hazardous wind farm data, and the representative physical features corresponding to the non-hazardous wind farm data, the method further includes: Marking the first data as the hazardous wind farm data, and marking at least part of the second data as non-hazardous wind farm data; The first data is the data corresponding to the hazardous moment in the Doppler velocity sampling data, the second data is the data in the Doppler velocity sampling data other than the first data, and the number of the hazardous wind field data is the same as the number of the non-hazardous wind field data.

4. The method according to any one of claims 1 to 3, wherein: The nonlinear residual network satisfies: Wherein, v is used to characterize the input Doppler velocity sampling data, z is used to characterize the representative physical characteristics of the input, ReLU() represents the rectified linear unit activation function, sigmoid[] represents the logarithmic probability function, {W1, W2, b1, b2} are the parameters of the nonlinear residual network, Indicates the output detection result.

5. The method according to claim 4, wherein: The loss function for training the nonlinear residual network satisfies: is the mean square error of the detection result, y i is the true label of the Doppler velocity sampling data of the i-th group, is the regularization term, and λ is the regularization coefficient.

6. An aviation hazard wind field detection device based on a nonlinear residual network, characterized by: include: A detection module is used to detect aircraft take-off and landing channels using radar, obtain Doppler velocity sampling data of the aircraft take-off and landing channels, and record the time when the aircraft encounters a hazardous wind field; An establishment module is used to establish a hazard feature library based on the second-order structure function of the Doppler velocity sampling data and traditional hazard assessment features corresponding to the Doppler velocity sampling data, wherein the traditional hazard assessment features include Doppler velocity average, turbulent energy dissipation rate, velocity fluctuation, F-factor and S-factor; An extraction module, configured to extract representative physical features from the hazard feature library; a training module, configured to train a nonlinear residual network based on hazardous wind field data, non-hazardous wind field data, representative physical features corresponding to the hazardous wind field data, and representative physical features corresponding to the non-hazardous wind field data, wherein the hazardous wind field data is data corresponding to the hazardous moment in the Doppler velocity sampling data, and the non-hazardous wind field data is data not corresponding to the hazardous moment in the Doppler velocity sampling data; A detection module, configured to perform aviation hazard wind field detection based on the nonlinear residual network to obtain a detection result; Wherein, the aviation hazard wind field detection device based on nonlinear residual network also includes: a calculation module, configured to calculate a second-order structure function of the Doppler velocity sampling data based on the Doppler velocity sampling data, and calculate a traditional hazard assessment feature corresponding to the Doppler velocity sampling data based on the Doppler velocity sampling data; The second-order structure function of the Doppler velocity sampling data satisfies: s(r)=mean([v(x)-v(x+r)] 2 ); Wherein, v(x) is used to represent the Doppler velocity sampling data detected by the radar at the position x of the aircraft take-off and landing channel, v(x+r) is used to represent the Doppler velocity sampling data detected by the radar at the position x+r of the aircraft take-off and landing channel, r represents the distance between x and x+r, and mean() represents the calculated average value; The Doppler velocity average value satisfies: M = mean(|v|); Wherein, v represents a set of Doppler velocity sampling data; The turbulent energy dissipation rate satisfies: Among them, C LOS is the Kolmogorov constant defined according to the Doppler velocity, κ1 and κ N is the wave number corresponding to the sampling distance, κ1=2π / r1,κ N =2π / r N ,σ v is the standard deviation of the Doppler velocity sampling data; The speed fluctuation satisfies: R = max(v) - min(v); Wherein, max(v) is the maximum value of the Doppler velocity sampling data, and min(v) is the minimum value of the Doppler velocity sampling data; The F-factor satisfies: Among them, u x is the tailwind speed along the aircraft take-off and landing channel, w is the vertical wind speed along the aircraft take-off and landing channel, V app is the aircraft airspeed, g is the acceleration due to gravity; The S-factor satisfies: Among them, △V i is the increment or decrement of the Doppler velocity sampling data, R i is the length of the corresponding slope in the aircraft take-off and landing channel, and the subscript i represents the i-th slope.

7. An electronic device comprising: A memory, a processor, and a program stored in the memory and executable on the processor; wherein the processor is configured to read the program in the memory to implement the steps of the aviation hazard wind field detection method based on a nonlinear residual network as described in any one of claims 1 to 5.

8. A readable storage medium for storing a program, characterized in that: When the program is executed by a processor, the steps of the aviation hazard wind field detection method based on a nonlinear residual network as claimed in any one of claims 1 to 5 are implemented.

Images