Anti-collision azimuth angle measurement method based on student psychology optimization algorithm

Abstract

The invention discloses an anti-collision azimuth angle measurement method based on a student psychology optimization algorithm, and the method comprises the steps: obtaining signal data transmitted by a radio frequency end, processing the signal data through employing a convolutional neural network, obtaining a sample label, defining a hardware link process from an antenna to a receiving and transmitting module, building an anti-collision angle measurement model through the combination of the sample label, and obtaining an anti-collision azimuth angle measurement model. A student teaching and learning optimization algorithm is adopted to carry out parameter configuration on an anti-collision angle measurement model, anti-collision azimuth angle measurement is carried out based on the anti-collision angle measurement model after parameter configuration, a set of reusable and high-precision anti-collision angle measurement model is formed by adopting an identification optimization thought, the multidimensionality of sample data is increased, and calculation errors are reduced.

Description

Technical Field

[0001] The invention relates to the technical field of airborne collision avoidance, and in particular to a collision avoidance azimuth measurement method based on a student psychology optimization algorithm. Background Art

[0002] With the rapid development of the aviation industry and the increasing dependence of human beings on aircraft, the number of aircraft in the sky is increasing, and the possibility of traffic accidents between aircraft is also increasing. How to ensure flight safety is receiving more and more attention. The TCAS (Traffic Collision Avoidance System, TCAS) system enables better communication and mutual monitoring between aircraft, effectively reducing the risk of collision between aircraft. After decades of continuous development, the TCAS system has now become an important guarantee for ensuring aircraft flight safety.

[0003] One of the important links to ensure the flight safety of aircraft is to monitor and track the aircraft, and to take a series of effective measures to achieve the purpose of collision avoidance by analyzing and processing the target aircraft's track.

[0004] Secondary radar can completely overcome the shortcomings of primary radar through the "inquiry-response" method. It sends inquiries to other airborne equipment through radar transmitters and directional antennas at a frequency of 1030MHz. After receiving valid inquiries, the airborne transponder uses a reply antenna to respond at a frequency of 1090MHz. Aircraft communicate with each other through secondary radar, and in this way, important information such as the other party's altitude, distance, direction and ICAO address can be obtained.

[0005] The antenna is mainly composed of omnidirectional radiation unit, directional radiation unit, directional beam network and corresponding structural parts. The omnidirectional radiation unit mainly realizes the reception and transmission of omnidirectional radio frequency signals, the directional radiation unit and the directional beam network realize directional beam synthesis, and the DC self-impedance detection circuit realizes the antenna online self-detection capability.

[0006] The directional feeding network can realize the power distribution of the input RF signal, ensure a specific phase relationship, and form the directional radiation of the antenna beam. On the other hand, it can provide a DC detection circuit. When the RF is fed to one of the ports, the power division and phase shift characteristics of the four 3dB branch-type directional couplers connected in series in the feeding network are used to form four feed sources with equal amplitudes and phases of 0°, -90°, -90°, and -180° to feed the directional radiation unit. This form of network can form four directional interrogation beams with a 3dB beam width of 90° in four quadrants when feeding J1, J2, J3, and J4 respectively. During DC detection, the ground resistance of the four RF connector input ends of the directional channel is designed to have different resistance values ​​to realize the system's detection of directional antennas. The resistance value of the J1 port is 1kΩ, the resistance value of the J2 port is 8kΩ, the resistance value of the J3 port is 4kΩ, and the resistance value of the J4 port is 2kΩ.

[0007] However, the azimuth information cannot be directly measured by the antenna. Instead, the azimuth of the corresponding target machine needs to be estimated using the OBA table based on the four amplitude values ​​(0°, -90°, -90°, and -180°) measured by the antenna. The OBA table involved in this method needs to be measured offline in advance, and due to the differences in antenna hardware, the OBA tables of different projects and batches of antennas need to be configured separately.

[0008] The current anti-collision azimuth angle measurement method uses a traversal query method, based on the minimum root mean square, to form an OBA table. The OBA table generated by this method cannot meet the scenario with high requirements for azimuth progress due to the small amount of measurement data and large root mean square error. Summary of the invention

[0009] The technical problem to be solved by the present invention is that the currently adopted method has little measurement data and large errors. The purpose is to provide an anti-collision azimuth angle measurement method based on a student psychology optimization algorithm, which adopts a convolutional neural network to process signal data, and constructs an anti-collision angle measurement model in combination with a defined hardware link process from an antenna to a transceiver module. The anti-collision angle measurement model is parameterized by a student teaching and learning optimization algorithm, and then the anti-collision angle measurement model is used to measure the anti-collision azimuth angle. By adopting the idea of ​​identification and optimization, a set of reusable and high-precision anti-collision angle measurement models is formed, which increases the multidimensionality of sample data and reduces calculation errors.

[0010] The present invention is achieved through the following technical solutions:

[0011] The present invention provides an anti-collision azimuth angle measurement method based on a student psychology optimization algorithm, comprising the following specific steps:

[0012] Obtain the signal data sent by the RF end, use a convolutional neural network to process the signal data, and obtain sample labels;

[0013] Define a hardware link process from antenna to transceiver module, and build an anti-collision angle measurement model based on sample labels;

[0014] The student teaching and learning optimization algorithm is used to configure the parameters of the anti-collision angle measurement model;

[0015] The anti-collision azimuth angle is measured based on the anti-collision angle measurement model after parameter configuration.

[0016] Furthermore, the signal data sent by the radio frequency end includes: A, C, S mode messages, amplitudes and antenna hardware parameters.

[0017] Furthermore, the convolutional neural network is used to process the signal data to obtain sample labels, including:

[0018] A neural network is used to perform weighted processing on the signal data, and the weighted signal data is aggregated into a group of weighted samples to obtain sample labels as data input for the anti-collision angle measurement model.

[0019] Furthermore, the weighted processing of the signal data by using a neural network includes:

[0020] Construct the input layer, convolution layer, pooling layer, fully connected layer and output layer, where:

[0021] The input layer is used to receive signal data sent by the radio frequency end;

[0022] The convolution layer is used to calculate the feature variables of the lth layer;

[0023] The pooling layer performs a pooling operation on the feature variables of the lth layer;

[0024] The fully connected layer is used to calculate the l-th layer feature vector;

[0025] The output layer is used to generate a prediction vector of a feature variable category according to Softmax regression to obtain a sample label.

[0026] Furthermore, the calculation of the characteristic variables of the lth layer includes:

[0027]

[0028] in, represents the characteristic variable of the lth layer, (j=1,…,N), and Respectively represent the offset of the convolution kernel and convolution layer; the operator symbol Represents the convolution operation; It represents the connection matrix between the convolutional layer and the feature variables of the previous layer, and the function f(x) represents the nonlinear activation function.

[0029] Furthermore, the pooling operation on the feature variables of the lth layer includes:

[0030]

[0031] Among them, p(x) is the pooling operation.

[0032] Furthermore, the calculation of the l-th layer feature vector includes:

[0033] x l =f(w l x l-1 +b l )(3)

[0034] Among them, w l and b l represents the weights and biases of the fully connected layer; the function f(x) represents the nonlinear activation function.

[0035] Furthermore, generating a prediction vector of a feature variable category includes:

[0036]

[0037] in, represents the sum of the exponential weights of all M categories, where M is the total number of categories, represents the i-th weight connecting the hidden layer and the output layer, L x is the number of hidden layers, L is the index of the layer, and L-1 represents the last hidden layer.

[0038] Furthermore, the process of defining a hardware link from an antenna to a transceiver module includes:

[0039] Build a hardware link, which includes an antenna, a four-terminal circulator, a limiter, a low-noise amplifier, a filter, and a low-pass filter;

[0040] The hardware link process from the antenna to the transceiver module includes:

[0041] The antenna receives TCAS / ADS-B signals through a directional channel;

[0042] TCAS / ADS-B signals enter the receiving channel after passing through a four-terminal circulator and a limiter;

[0043] The TCAS / ADS-B signal entering the receiving channel is amplified by a low-noise amplifier;

[0044] The amplified TCAS / ADS-B signal enters the dielectric filter for processing to obtain a processed TCAS / ADS-B signal.

[0045] Furthermore, the use of the student teaching and learning optimization algorithm to configure parameters of the anti-collision angle measurement model includes:

[0046] Based on student teaching and learning optimization algorithm:

[0047] Obtain the anti-collision angle measurement configuration parameters to be identified for the best student, good student, average student, student trying to improve randomly, and the jth randomly selected student;

[0048] According to the anti-collision angle measurement configuration parameters to be identified of the best student, the good student, the ordinary student, the student trying to randomly improve and the jth randomly selected student, the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the ordinary anti-collision angle measurement configuration parameters to be identified and the anti-collision angle measurement configuration parameters to be identified of the student trying to randomly improve are obtained;

[0049] According to the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the anti-collision angle measurement configuration parameters to be identified of the ordinary student and the anti-collision angle measurement configuration parameters to be identified of the student who attempts random improvement, the optimal configuration parameters after random improvement are obtained.

[0050] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0051] The present invention aims to perform weighted classification processing on antenna hardware parameters and A, C, and S message information through an algorithm that integrates big data and neural networks, and then adopts the idea of ​​identification to iteratively calculate the designed anti-collision angle measurement model to form a set of logical models that can perform anti-collision angle measurement;

[0052] The anti-collision angle measurement model designed by the present invention involves the hardware parameters and message information of the antenna, which can be better mapped to the azimuth angle and improve the accuracy of anti-collision angle measurement;

[0053] The anti-collision angle measurement method designed in the present invention develops an anti-collision angle measurement model based on big data, which can better reflect the azimuth characteristics of the antenna, can better adapt to different batches and different projects, and improves versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0055] Figure 1 The figure is a flow chart of the anti-collision angle measurement accuracy design method in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0057] As a possible implementation, Figure 1 As shown, this embodiment provides an anti-collision azimuth angle measurement method based on a student psychology optimization algorithm, including: data integration and classification, anti-collision angle measurement model construction, and configuration parameter identification of the anti-collision angle measurement model. Specifically, signal data sent by the radio frequency end is obtained, and the signal data is processed by a convolutional neural network to obtain a sample label; a hardware link process from an antenna to a transceiver module is defined, and an anti-collision angle measurement model is constructed in combination with the sample label; the anti-collision angle measurement model is configured with parameters using a student teaching and learning optimization algorithm; and anti-collision azimuth angle measurement is performed based on the anti-collision angle measurement model after parameter configuration.

[0058] In this embodiment:

[0059] It is necessary to collect the anti-collision angle measurement data generated by multiple projects and different batches, and summarize and organize the data in sequence according to feature classification to form an anti-collision angle measurement database;

[0060] It is necessary to design a feasible anti-collision angle measurement model, which should combine the hardware link process from antenna to transceiver module;

[0061] It is necessary to develop corresponding intelligent optimization software in combination with the anti-collision angle measurement database and the anti-collision angle measurement model, and modulate the configuration parameters of the anti-collision angle measurement model according to the azimuth error;

[0062] It is necessary to add the configuration reading and storage functions of the antenna hardware characteristics so that the anti-collision angle measurement model can automatically adapt the configuration parameters according to the antenna hardware characteristics, thereby improving the versatility of the anti-collision angle measurement model;

[0063] First, for data integration and classification, a convolutional neural network is used to process the amplitude, A, C, S mode messages, and antenna hardware parameter information sent by the RF end. The neural network is used to weight the data and summarize them into a set of weighted samples. These samples are used as data input for the anti-collision angle measurement model.

[0064] A convolutional neural network can be understood as a neural network with many hidden layers. According to the positions of different layers, the neural network layers in a deep neural network can be divided into input layer, convolution layer, pooling layer, fully connected layer and output layer. The layers are fully connected.

[0065] in:

[0066] The input layer is used to receive multi-dimensional signal data sent by the RF end;

[0067] The convolution layer is used to calculate the feature variables of the first layer. The features extracted by convolution will pay more attention to the local area. Each neuron does not need to perceive the global image, but only needs to perceive the local area. By integrating the local information at a higher layer, all information can be obtained, thereby marking different weighted values ​​for surveillance data of different threat levels.

[0068] The pooling layer performs a pooling operation on the feature variables of the first layer. The pooling layer is usually arranged after the convolution layer. By performing a pooling operation on the local area of ​​the feature variables, the features have a certain spatial invariance.

[0069] The fully connected layer is used to calculate the feature vector of the first layer, specifically to connect all the features and send the output value to the classifier.

[0070] Thus, data classification can be realized at the monitoring end;

[0071] Specifically,

[0072] The function of the convolution layer is to extract features using convolution operations. The more convolution layers there are, the stronger the feature expression ability is. If the lth layer is a convolution layer, the feature variable of this layer can be calculated by the following formula (j=1,…,N):

[0073]

[0074] in: and Respectively represent the offset of the convolution kernel and convolution layer; the operator symbol Represents the convolution operation; Represents the connection matrix between the convolutional layer and the feature variables of the previous layer. If is 1, then the characteristic variable With characteristic variables Related, if is 0, then the characteristic variable With characteristic variables No association; the function f(x) represents a nonlinear activation function.

[0075] The pooling layer is usually set after the convolution layer. By performing pooling operations on the local area of ​​the feature variable, the feature has a certain spatial invariance. If the 1+1 layer is a pooling layer, then the feature variable of this layer can be calculated in the following way (j=1,…,N):

[0076]

[0077] Where: p(x) is the pooling operation. Commonly used pooling operations include mean pooling and maximum pooling.

[0078] The fully connected layer is located after feature extraction and connects all neurons in the previous layer to each neuron in the current layer. The fully connected layer maps high-level features in a targeted manner according to the specific tasks of the output layer. If the first layer is a fully connected layer and the previous layer is also a fully connected layer, then the first layer feature vector x l The calculation method is:

[0079] x l =f(w l x l-1 +b l )(3)

[0080] Among them, w l and b l represents the weights and offsets of the fully connected layer; the function f(x) represents the nonlinear activation function. If the first layer is a fully connected layer and the previous layer is a convolutional layer or a pooling layer, then there are two ways to calculate the feature vector of the first layer.

[0081] The fully connected layer is regarded as a convolutional layer with a feature variable size of 1×1. The number of feature variables is equal to the number of neurons in the fully connected layer. Then the calculation method of each feature variable is:

[0082]

[0083] The form of the output layer is task-oriented. If the convolutional neural network is used as a classifier, the output layer uses Softmax regression to generate a prediction vector y = (yi, ..., yM) T of the feature variable category. The calculation method is:

[0084]

[0085] in: represents the sum of the exponential weights of all M categories, where M is the total number of categories, represents the i-th weight connecting the hidden layer and the output layer, L x is the number of hidden layers, L is the index of the layer, and L-1 represents the last hidden layer.

[0086] Secondly, in the construction of the anti-collision angle measurement model, that is, the offline development stage, it is necessary to define a dynamic model that can reflect the signal transmission process from the antenna to the transceiver module. The model is associated with the hardware link and amplitude information. The establishment of this model is the key to the offline development stage, and they provide important theoretical support and technical basis for subsequent real-time control and simulation. In the hardware link designed, the antenna directional channel receives the TCAS / ADS-B signal through a four-terminal circulator and enters the receiving channel after limiting. After LNA amplification, it enters the front-end dielectric filter to select the received signal, complete the suppression of out-of-band signals, and simultaneously realize the suppression of image frequency and intermediate frequency, and realize out-of-band intermodulation together with the LTCC low-pass filter behind; then sampling is performed to realize the TCAS / ADS-B signal amplitude output and video output.

[0087] Finally, the configuration parameters of the anti-collision angle measurement model are identified, and the parameters of the anti-collision angle measurement model are configured using the student teaching and learning optimization algorithm, including:

[0088] Based on student teaching and learning optimization algorithm:

[0089] Obtain the anti-collision angle measurement configuration parameters to be identified of the best student, the good student, the ordinary student, the student trying to improve randomly and the jth randomly selected student, and obtain the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the ordinary anti-collision angle measurement configuration parameters to be identified and the anti-collision angle measurement configuration parameters to be identified of the student trying to improve randomly;

[0090] According to the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the anti-collision angle measurement configuration parameters to be identified of the ordinary student and the anti-collision angle measurement configuration parameters to be identified of the student who attempts random improvement, the optimal configuration parameters after random improvement are obtained.

[0091] For SPBO (Student Teaching and Learning Optimization Algorithm), a heuristic optimization algorithm is developed based on the psychology of students in the class. The algorithm is divided into four iterative modes: the best student, good student, average student, and students who try to improve randomly. The effort made by students in a subject depends on the interest obtained in the subject, which depends on the student psychology and may vary from student to student and subject to subject. Based on this psychology, in the proposed SPBO algorithm, except for the best student, the subject selection of the other three types of students is considered to be a random process.

[0092] Applying the SPBO algorithm to the configuration parameter identification of the anti-collision angle measurement model can recursively calculate the expected cost of each state.

[0093] The best students contribute more to reaching the best solution and often work harder than any randomly selected student;

[0094] A good student will try to work harder on any subject in order to improve his overall performance. Some students will try to work as hard or harder than the best students;

[0095] Some students try to study harder than the average student in the class and try to emulate the best students;

[0096] An average student is a student of average level in the class. While studying the subject at the average level of the class, the student will try to put more effort in other subjects.

[0097] The specific process of SPBO algorithm is as follows:

[0098] Best Student:

[0099] θ best new,m =θ best,m +(-1) k ×rand×(θ best,m -θ j,m ) (6)

[0100] Good students:

[0101] θ newi,m =θ best,m +rand×(θ best,m -θ i,m ) (7)

[0102] Students who improved randomly:

[0103] θ newi,m =θ i,m +rand×[(θ best,m -θ i,m )+(θ i,m -θ mean,m )] (8)

[0104] Ordinary students:

[0105] θ newi,m =θ i,m +rand×(θ mean,m -θ i,m ) (9)

[0106] Wherein, in formula (6), θbest,m represents the best student, θj,m represents the anti-collision angle measurement configuration parameters to be identified of the jth randomly selected student, θnew,best,m represents the updated anti-collision angle measurement configuration parameters to be identified of the best student in the next iteration, k represents a randomly selected 1 or 2, which is used to determine the sign of the update rule, and rand represents a random number in [0,1], which is used to determine the update amplitude; formulas (7)-(9) correspond to the parameter interpretation of formula (6), respectively, corresponding to the updated anti-collision angle measurement configuration parameters to be identified of the good student, the randomly improved student, and the ordinary student in the next iteration.

[0107] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for measuring anti-collision azimuth angle based on student psychology optimization algorithm, characterized in that: The specific steps include: Obtain the signal data sent by the RF end, use a convolutional neural network to process the signal data, and obtain sample labels; Define a hardware link process from antenna to transceiver module, and build an anti-collision angle measurement model based on sample labels; The student teaching and learning optimization algorithm is used to configure the parameters of the anti-collision angle measurement model; The anti-collision azimuth angle is measured based on the anti-collision angle measurement model after parameter configuration.

2. The anti-collision azimuth angle measurement method based on the student psychology optimization algorithm according to claim 1 is characterized in that: The signal data sent by the radio frequency end includes: A, C, S mode messages, amplitudes and antenna hardware parameters.

3. The anti-collision azimuth angle measurement method based on the student psychology optimization algorithm according to claim 1 is characterized in that: The convolutional neural network is used to process the signal data to obtain sample labels, including: A neural network is used to perform weighted processing on the signal data, and the weighted signal data is aggregated into a group of weighted samples to obtain sample labels as data input for the anti-collision angle measurement model.

4. The anti-collision azimuth angle measurement method based on the student psychology optimization algorithm according to claim 3 is characterized in that: The method of using a neural network to perform weighted processing on signal data includes: Construct the input layer, convolution layer, pooling layer, fully connected layer and output layer, where: The convolution layer is used to calculate the feature variables of the lth layer; The pooling layer performs a pooling operation on the feature variables of the lth layer; The fully connected layer is used to calculate the l-th layer feature vector; The output layer is used to generate a prediction vector of a feature variable category according to Softmax regression to obtain a sample label.

5. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 1 is characterized in that: The calculating of the characteristic variables of the lth layer includes: in, represents the characteristic variable of the lth layer, (j=1,…,N), and Respectively represent the offset of the convolution kernel and convolution layer; the operator symbol Represents the convolution operation; It represents the connection matrix between the convolutional layer and the feature variables of the previous layer, and the function f(x) represents the nonlinear activation function.

6. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 5 is characterized in that: The pooling operation on the feature variables of the lth layer includes: Among them, p(x) is the pooling operation.

7. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 6 is characterized in that: The step of calculating the l-th layer feature vector comprises: x l =f(w l x l-1 +b l )(3) Among them, w l and b l represents the weights and biases of the fully connected layer; the function f(x) represents the nonlinear activation function.

8. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 7 is characterized in that: The step of generating a prediction vector of a feature variable category includes: in, represents the sum of the exponential weights of all M categories, where M is the total number of categories, represents the i-th weight connecting the hidden layer and the output layer, L x is the number of hidden layers, L is the index of the layer, and L-1 represents the last hidden layer.

9. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 1 is characterized in that: The process of defining a hardware link from an antenna to a transceiver module includes: Build a hardware link, which includes an antenna, a four-terminal circulator, a limiter, a low-noise amplifier, a filter, and a low-pass filter; The hardware link process from the antenna to the transceiver module includes: The antenna receives TCAS / ADS-B signals through a directional channel; TCAS / ADS-B signals enter the receiving channel after passing through a four-terminal circulator and a limiter; The TCAS / ADS-B signal entering the receiving channel is amplified by a low noise amplifier; The amplified TCAS / ADS-B signal enters the dielectric filter for processing to obtain a processed TCAS / ADS-B signal.

10. The anti-collision azimuth angle measurement method based on student psychology optimization algorithm according to claim 1 is characterized in that: The method of using the student teaching and learning optimization algorithm to configure parameters of the anti-collision angle measurement model includes: Based on student teaching and learning optimization algorithm: Obtain the anti-collision angle measurement configuration parameters to be identified for the best student, good student, average student, student trying to improve randomly, and the jth randomly selected student; According to the anti-collision angle measurement configuration parameters to be identified of the best student, the good student, the ordinary student, the student trying to randomly improve and the jth randomly selected student, the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the ordinary anti-collision angle measurement configuration parameters to be identified and the anti-collision angle measurement configuration parameters to be identified of the student trying to randomly improve are obtained; According to the anti-collision angle measurement configuration parameters to be identified of the best student, the anti-collision angle measurement configuration parameters to be identified of the good student, the anti-collision angle measurement configuration parameters to be identified of the ordinary student and the anti-collision angle measurement configuration parameters to be identified of the student who attempts random improvement, the optimal configuration parameters after random improvement are obtained.

Images