A method for predicting a flight trajectory of an aircraft

Abstract

The present disclosure relates to a flight trajectory prediction method of an aircraft, and relates to the technical field of hypersonic aircrafts, and the method comprises the following steps: determining a raw trajectory database of a hypersonic aircraft under different initial states and different flight control laws based on a motion equation; training a network model to be trained based on the raw trajectory database to obtain a current trajectory prediction model; determining a trajectory prediction model to be migrated from the current trajectory prediction model according to a target flight trajectory of an aircraft to be predicted and a raw flight trajectory in the raw trajectory database, and migrating model parameters in the trajectory prediction model to be migrated to a trajectory prediction model to be corrected; performing parameter correction on the trajectory prediction model to be corrected based on the target flight trajectory to obtain a target trajectory prediction model, and predicting a flight trajectory of the aircraft to be predicted based on the target trajectory prediction model. The present disclosure improves the accuracy of the predicted flight estimation.

Description

TECHNICAL FIELD

[0001] The embodiment of the present disclosure relates to the technical field of hypersonic aircraft, in particular to a flight trajectory prediction method of an aircraft. BACKGROUND

[0002] At present, the trajectory prediction based on machine learning is to train a model in an offline training manner under the condition that the training set and the test set data are in the same distribution; meanwhile, in the process of actual application, the fitting error of the model can be taken as an evaluation standard, the neural network prediction model is trained through a large amount of observation data, and a better prediction result is obtained in the test set by using different neural networks.

[0003] However, since the observation target (i.e. the aircraft to be predicted) is often non-cooperative, the important information such as the mass, volume and flight control law of the aircraft to be predicted cannot be obtained by the defense party, the conventional machine learning trajectory prediction algorithm needs to train a model from scratch, which requires a large amount of observation data, and thus the training efficiency of the model is low; meanwhile, the accuracy of the trajectory prediction result of the obtained model is also low.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and thus can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present disclosure is to provide a flight trajectory prediction method of an aircraft, thereby at least partially overcoming the problem of low training efficiency of the model and low accuracy of the trajectory prediction result of the model caused by the limitations and defects of the related art.

[0006] According to one aspect of the present disclosure, a flight trajectory prediction method of an aircraft is provided, comprising:

[0007] establishing a motion equation of a hypersonic aircraft, and determining an original trajectory database of the hypersonic aircraft under different initial states and different flight control laws based on the motion equation;

[0008] training a network model to be trained based on the original trajectory database to obtain a current trajectory prediction model;

[0009] determining a trajectory prediction model to be migrated from the current trajectory prediction model according to a target flight trajectory of an aircraft to be predicted and an original flight trajectory in the original trajectory database, and migrating the model parameters in the trajectory prediction model to be migrated to a trajectory prediction model to be corrected;

[0010] Parameter correct the trajectory prediction model to be corrected based on the target flight trajectory, obtain a target trajectory prediction model, and predict the flight trajectory of the to-be-predicted aircraft based on the target trajectory prediction model.

[0011] In an exemplary embodiment of the present disclosure, a motion equation of a hypersonic aircraft is established, including:

[0012] The flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude, and the latitude of the hypersonic aircraft are obtained.

[0013] In the ballistic coordinate system, a three-degree-of-freedom motion equation of the hypersonic aircraft is established according to the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude, and the latitude.

[0014] In an exemplary embodiment of the present disclosure, an original ballistic database of a hypersonic aircraft under different initial states and different flight control laws is determined based on the motion equation, including:

[0015] Different initial state variables of the hypersonic aircraft are determined based on the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude, and the latitude.

[0016] Different flight control laws of the hypersonic aircraft are determined according to the angle of attack control instruction and the roll angle control instruction of the hypersonic aircraft.

[0017] The different initial state variables and the different flight control laws are input into the three-degree-of-freedom motion equation to obtain the original flight trajectory under the different initial state variables and the different flight control laws.

[0018] The original ballistic database is generated according to the original flight trajectory and the initial state variables and the control laws corresponding to the original flight trajectory.

[0019] In an exemplary embodiment of the present disclosure, a current trajectory prediction model is obtained by training a to-be-trained network model based on the original ballistic database, including:

[0020] The initial state variables and the control laws corresponding to the original flight trajectory in the original ballistic database are input into the to-be-trained network model to obtain a current prediction result.

[0021] A current loss function is constructed according to the current prediction result and the original flight trajectory, and parameters in the to-be-trained network model are adjusted based on the current loss function to obtain a current trajectory prediction model.

[0022] In an example embodiment of the present disclosure, determining the trajectory prediction model to be migrated from the current trajectory prediction model according to the target flight trajectory of the aircraft to be predicted and the original flight trajectory in the original trajectory database comprises:

[0023] Obtaining the target flight trajectory of the aircraft to be predicted, and performing noise reduction processing on the target flight trajectory to obtain a flight trajectory after noise reduction processing;

[0024] Determining a first trajectory sequence of the flight trajectory after noise reduction processing, and determining a second trajectory sequence of the original flight trajectory in the original trajectory database;

[0025] Calculating the similarity of the first trajectory sequence and the second trajectory sequence, and determining the trajectory prediction model to be migrated from the current trajectory prediction model according to the similarity.

[0026] In an example embodiment of the present disclosure, calculating the similarity of the first trajectory sequence and the second trajectory sequence, and determining the trajectory prediction model to be migrated from the current trajectory prediction model according to the similarity comprises:

[0027] Calculating the Euclidean distance between the first trajectory sequence and the second trajectory sequence, and determining the similarity of the first trajectory sequence and the second trajectory sequence according to the Euclidean distance;

[0028] Sorting the original flight trajectories based on the similarity of the first trajectory sequence and the second trajectory sequence, and extracting the original flight trajectory with the smallest similarity from the original flight trajectories based on the trajectory sorting result;

[0029] Taking the current trajectory prediction model corresponding to the original flight trajectory with the smallest similarity as the trajectory prediction model to be migrated.

[0030] In an example embodiment of the present disclosure, migrating the model parameters in the trajectory prediction model to be migrated to the trajectory prediction model to be corrected comprises:

[0031] Extracting the model parameters in the trajectory prediction model to be migrated, and performing parameter initialization processing on the model parameters in the trajectory prediction model to be corrected based on the extracted model parameters to realize the parameter migration processing.

[0032] In an example embodiment of the present disclosure, performing parameter correction on the trajectory prediction model to be corrected based on the target flight trajectory to obtain a target trajectory prediction model comprises:

[0033] Obtaining the sample data of the aircraft to be predicted observed, and inputting the sample data into the trajectory prediction model to be corrected to obtain a target prediction result;

[0034] Based on the target prediction results and the target flight trajectory, a target loss function is constructed, and the parameters of the trajectory prediction model to be corrected are corrected based on the target loss function to obtain the target trajectory prediction model.

[0035] In one exemplary embodiment of this disclosure, the trajectory prediction model to be corrected includes a convolutional neural network model and a bidirectional long short-term memory network model;

[0036] Specifically, the trajectory prediction model to be corrected is parameter-corrected based on the target loss function to obtain the target trajectory prediction model, including:

[0037] The weights of the convolutional neural network are fixed, and the weights and biases of the bidirectional long short-term memory network model are fine-tuned based on the target loss function to obtain the target trajectory prediction model.

[0038] In the process of fine-tuning the weights and biases of the bidirectional long short-term memory network model, the selected optimizer includes the Adam optimizer with adaptive learning rate adjustment.

[0039] In one exemplary embodiment of this disclosure, predicting the flight trajectory of the aircraft to be predicted based on the target trajectory prediction model includes:

[0040] The flight speed, trajectory inclination angle, trajectory deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude of the aircraft to be predicted are obtained.

[0041] The model input vector is constructed based on the flight speed, trajectory tilt angle, trajectory deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude of the aircraft to be predicted.

[0042] The model input vector is input into the target trajectory prediction model to obtain the flight trajectory prediction result of the aircraft to be predicted.

[0043] The flight trajectory prediction method of the aircraft provided in the embodiments of the present disclosure, on the one hand, can establish the motion equation of the hypersonic aircraft, determine the original trajectory database of the hypersonic aircraft under different initial states and different flight control laws based on the motion equation, then train the network model to be trained based on the original trajectory database to obtain the current trajectory prediction model, then determine the trajectory prediction model to be migrated from the current trajectory prediction model according to the target flight trajectory of the aircraft to be predicted and the original flight trajectory in the original trajectory database, and migrate the model parameters in the trajectory prediction model to be migrated to the trajectory prediction model to be corrected, finally correct the parameters of the trajectory prediction model to be corrected based on the target flight trajectory to obtain the target trajectory prediction model, and predict the flight trajectory of the aircraft to be predicted based on the target trajectory prediction model, thereby improving the accuracy of the predicted flight trajectory and solving the problem of low accuracy of the model trajectory prediction result in the prior art; on the other hand, the model parameters in the trajectory prediction model to be migrated can be migrated to the trajectory prediction model to be corrected, and finally the parameters of the trajectory prediction model to be corrected are corrected based on the target flight trajectory to obtain the target trajectory prediction model, without training the trajectory prediction model to be corrected from scratch, thereby improving the training efficiency of the model.

[0044] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings, which are incorporated into and form part of the specification, illustrate an embodiment consistent with the present disclosure and, together with the specification, serve to explain the principles of the present disclosure. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0046] Figure 1 A flowchart of a flight trajectory prediction method of an aircraft according to an example embodiment of the present disclosure is schematically shown.

[0047] Figure 2 An example diagram of a model structure of a CNN-BiLSTM according to an example embodiment of the present disclosure is schematically shown.

[0048] Figure 3 An example diagram of a framework of the implementation principle of a CNN-BiLSTM according to an example embodiment of the present disclosure is schematically shown.

[0049] Figure 4 An example diagram of the structure of a 1DCNN according to an example embodiment of the present disclosure is schematically shown.

[0050] Figure 5 An example diagram of a structure of an LSTM is schematically shown according to an example embodiment of the present disclosure.

[0051] Figure 6 An example diagram of a structure of a BiLSTM is schematically shown according to an example embodiment of the present disclosure.

[0052] Figure 7 An example diagram of a specific generation process of a raw ballistic database is schematically shown according to an example embodiment of the present disclosure.

[0053] Figure 8 An example diagram of performing aircraft trajectory prediction based on a 1D CNN-BiLSTM model is schematically shown according to an example embodiment of the present disclosure.

[0054] Figure 9 A block diagram of an aircraft flight trajectory prediction apparatus is schematically shown according to an example embodiment of the present disclosure.

[0055] Figure 10 An electronic device for implementing an aircraft flight trajectory prediction method is schematically shown according to an example embodiment of the present disclosure. DETAILED DESCRIPTION

[0056] Example embodiments now will be described more fully hereinafter with reference to the accompanying drawings. Example embodiments, however, can be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example embodiments to those skilled in the art. Like reference numerals refer to like elements throughout the description. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; thus, the scope of the example embodiments should be interpreted broadly, as being limited only by the claims and any equivalents thereof. The example embodiments described herein are not intended to be limited to the specific forms set forth herein, but on the contrary, are intended to cover such alternatives, modifications, and equivalents as can be within the scope of the example embodiments. The examples described herein are intended to be illustrative of the example embodiments and should not be construed as limiting the scope of the example embodiments.

[0057] Furthermore, the accompanying drawings are intended to be illustrative only and are not necessarily drawn to scale. Identical reference numerals denote identical or similar parts throughout the entire specification. Some of the block diagrams shown in the drawings are functional entities that do not necessarily have to correspond to physically or logically independent entities. These functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0058] In recent years, a lot of achievements have been made in the trajectory prediction of hypersonic vehicles. At present, the trajectory prediction based on machine learning mainly considers the case that the training set and the test machine data are in the same distribution, the model is trained offline, the fitting error of the model is taken as the evaluation standard, a large amount of observation data is used to train the neural network prediction model, and a better prediction result is obtained by using different neural networks in the test set. However, since the observation target is often non-cooperative, the defense party cannot obtain important information such as the mass, volume and flight control law of the target vehicle, and the conventional machine learning trajectory prediction algorithm needs to train a model from scratch, which requires a large amount of observation data and has poor real-time prediction performance.

[0059] In view of this, a 1DCNN-BiLSTM transfer learning trajectory prediction model is proposed in this paper. For the non-cooperative target trajectory prediction problem, a 1DCNN-BiLSTM prediction model library is trained in the form of a trajectory database, and under the condition of small sample observation data, the similar trajectory network parameters are fine-tuned to realize fast and high-precision prediction. The flight trajectory prediction method of the aircraft provided by the example embodiment of the present disclosure can maximize the value of the offline trajectory database, and can quickly predict the target trajectory with high precision under the condition of small sample observation, so as to improve the response speed of the anti-missile defense system.

[0060] In an example embodiment, the flight trajectory prediction method of the aircraft is first provided in the example embodiment, which can run on a terminal device, a server, a server cluster or a cloud server, etc. Of course, those skilled in the art can also run the method of the present disclosure on other platforms according to the needs, which is not specially limited in the example embodiment. Specifically, referring to Figure 1 The flight trajectory prediction method of the aircraft can include the following steps:

[0061] Step S110. Establishing the motion equation of the hypersonic vehicle, and determining the original trajectory database of the hypersonic vehicle under different initial states and different flight control laws based on the motion equation;

[0062] Step S120. Training the network model to be trained based on the original trajectory database to obtain a current trajectory prediction model;

[0063] Step S130. According to the target flight trajectory of the aircraft to be predicted and the original flight trajectory in the original trajectory database, determining a trajectory prediction model to be transferred from the current trajectory prediction model, and transferring the model parameters in the trajectory prediction model to be transferred to the trajectory prediction model to be corrected;

[0064] Step S140. Parameter correcting the trajectory prediction model to be corrected based on the target flight trajectory, obtaining a target trajectory prediction model, and predicting the flight trajectory of the to-be-predicted aircraft based on the target trajectory prediction model.

[0065] In the flight trajectory prediction method of the aircraft, on the one hand, the motion equation of the hypersonic aircraft can be established, and the original trajectory database of the hypersonic aircraft under different initial states and different flight control laws can be determined based on the motion equation; then the to-be-trained network model is trained based on the original trajectory database to obtain a current trajectory prediction model; then the to-be-migrated trajectory prediction model is determined from the current trajectory prediction model according to the target flight trajectory of the to-be-predicted aircraft and the original flight trajectory in the original trajectory database, and the model parameters in the to-be-migrated trajectory prediction model are migrated to the trajectory prediction model to be corrected; finally, the trajectory prediction model to be corrected is parameter corrected based on the target flight trajectory to obtain a target trajectory prediction model, and the flight trajectory of the to-be-predicted aircraft is predicted based on the target trajectory prediction model, thereby improving the accuracy of the predicted flight trajectory and solving the problem of low accuracy of the model trajectory prediction result in the prior art. On the other hand, the model parameters in the to-be-migrated trajectory prediction model can be migrated to the trajectory prediction model to be corrected; finally, the trajectory prediction model to be corrected is parameter corrected based on the target flight trajectory to obtain a target trajectory prediction model, without the need to train the trajectory prediction model to be corrected from scratch, thereby improving the training efficiency of the model.

[0066] Hereinafter, the flight trajectory prediction method of the aircraft according to the example embodiments of the present disclosure will be explained and described in detail with reference to the accompanying drawings.

[0067] Firstly, the technical implementation principle of the example embodiments of the present disclosure is explained and described. Specifically, in order to realize the prediction of the flight trajectory of the to-be-predicted aircraft, the flight trajectory prediction method of the aircraft described in the example embodiments of the present disclosure needs to establish an offline trajectory database of the aircraft in the actual application process. Specifically, in the process of constructing the offline trajectory database (i.e., the original trajectory database or the source domain trajectory database), firstly, the forces acting on the aircraft need to be considered, and the three-degree-of-freedom motion equation of the aircraft is established in the trajectory coordinate system. Then, the angle of attack-velocity profile is set, the flight state of the aircraft is divided into two segments of the initial diving segment and the descending segment, and the function of the angle of attack changing with the speed is obtained. Then, the heat flow constraint, the overload constraint and the dynamic pressure constraint that the aircraft needs to satisfy in the flight process are established. Finally, the flight initial conditions and the angle of attack control instructions and the roll angle control instructions are given, and the simulation method of the Monte Carlo trajectory database is constructed. Secondly, the 1DCNN-BiLSTM network and the transfer learning framework are established. Specifically, the aircraft trajectory prediction model framework is established, which mainly consists of two parts: (1) Generation of the source domain trajectory database. The offline trajectory database is simulated by Monte Carlo, and the neural network is trained for the trajectories in the database. After multiple iterations, a high-precision pre-training model library is obtained. (2) Target domain trajectory prediction. The network parameters of the most similar trajectory in the source domain are transferred to the target domain, and the trajectory of the target aircraft is predicted after network fine-tuning, and the future trajectory of the target is calculated. Further, the trajectory prediction is performed based on the 1DCNN-BiLSTM and the transfer learning. Specifically, the source domain trajectory database is generated, the target domain data is processed, the similarity is selected, the trajectory prediction is performed after the network fine-tuning, and the prediction model is evaluated. Based on the flight trajectory prediction method of the aircraft described above, the similar trajectory network parameters can be fine-tuned to realize fast and high-precision prediction under the condition of small sample observation data; at the same time, the value of the offline trajectory database can be maximized.

[0068] Secondly, the network model related to the example embodiments of the present disclosure is explained and described. Specifically, the network model to be trained and the trajectory prediction model to be corrected related to the example embodiments of the present disclosure are realized by CNN-BiLSTM. The specific model structure example diagram of CNN-BiLSTM can be referred to Figure 2 The implementation principle framework example diagram of CNN-BiLSTM can be referred to Figure 3As shown; further, the CNN described here refers to Convolutional Neural Networks, and the BiLSTM described here refers to Bidirectional Long Short Term Memory; further, the CNN described here can be 1DCNN, or of course 2DCNN or 3DCNN, and this example does not impose any special restrictions on this.

[0069] In one example embodiment, the 1DCNN network described herein can be used to extract feature information from the input vector input to the model. Specifically, in practical applications, convolutional neural networks (CNNs) can map simple data to a high-dimensional space through convolutional layers and pooling layers to extract feature information. It should be noted that while CNNs are structurally similar to multilayer perceptrons and can effectively address the problem of low prediction accuracy, multilayer perceptrons suffer from excessively large parameter sizes due to their fully connected nature. Therefore, to avoid this problem, convolutional neural networks have been improved by replacing fully connected connections with sparse connections to link adjacent neurons and performing pooling operations on the processed features. Furthermore, this example embodiment uses a 1DCNN for illustration. A 1DCNN is a type of CNN used for feature extraction and classification of one-dimensional data. The 1DCNN described herein can be composed of an input layer 401, a one-dimensional convolutional layer 402, a pooling layer 403, and an output layer 404. A specific structural example diagram can be found in the provided diagram. Figure 4 As shown.

[0070] Furthermore, in practical applications, the input data x i When the convolutional layer is calculated, the formula can be shown in formula (1) below:

[0071]

[0072] The specific formula for calculating the pooling layer is shown in formula (2) below:

[0073]

[0074] Where f is the activation function. Let be the output of the i-th neuron in the l-th layer. The output of the i-th neuron in layer l-1; r is the convolution kernel compensation; j represents the sparse connection between neurons in two layers; These are the weights of the corresponding neurons; This represents the bias of the corresponding neuron. This is the (l-1)th characteristic matrix.

[0075] In an example embodiment, the BiLSTM network described herein can be used for trajectory prediction based on the features extracted by the CNN. Specifically, the LSTM network is improved from the recurrent neural network (RNN), and by adding a gating mechanism, the historical information is recorded, and the forgetting problem of the traditional RNN network to the past data is solved. However, the single-layer LSTM network only considers the forward historical information in the training process, and lacks the subsequent historical information, so its accuracy is still insufficient in long-term prediction, and the BiLSTM network is composed of two LSTM networks in the forward and reverse directions. The forward network obtains the forward historical information of the input sequence, and the backward network inputs the future information of the sequence, and further mines the overall time sequence information of the trajectory. The unit structure of the LSTM can be as shown in Figure 5 Further, in the structure example diagram shown in Figure 5 The LSTM network uses a more complex internal neural network to save and use sequence information, and alleviates the gradient disappearance or gradient explosion problem of the RNN when processing long sequences. The LSTM unit has three gate structures in total:

[0076] The first gate is the forget gate structure. The forget gate reads the hidden state at the last moment and the current input, and after the simoid function, the current input of the LSTM unit and the hidden state H t of the last LSTM unit are discriminated, and part of the features are selectively forgotten, and the useful features are retained;

[0077] The second is the input gate structure. The input gate selectively records the current input and the hidden state H t of the last LSTM unit into the cell state by the Sigmoid function, and updates the cell information C t in the LSTM unit by using the Tanh layer combined with the data information retained by the forget gate;

[0078] The third is the output gate structure. The output gate selectively outputs the C t data information in the current unit by the Sigmoid layer, and then multiplies the cell state information after the Tanh layer normalization processing and the output of the Sigmoid layer to obtain the final output H t of the cell. The final state is only the output part determined by the output gate.

[0079] Among them, the functions used by the LSTM network are Sigmoid, Tanh, Relu, etc. The forward calculation process of the network can be as shown in the following formula (3):

[0080]

[0081] Wherein, i represents an input gate, f represents a forget gate, o represents an output gate; H t , t-1 represents a current output and a previous time step output, C t , t-1 is a unit information of a previous time step, is newly generated information, W xi ,W xf ,W xo ,W hi ,W hO ,W hc and b i ,b f ,b O ,b c are neural network weight and bias vectors corresponding to each gate structure respectively. Wherein, the cell information C t and H t are updated in the form of element-wise multiplication within a vector; sigma represents an activation function of a neural network, and a suitable activation function is selected to give the LSTM network better nonlinear fitting ability.

[0082] Further, the structure of the BiLSTM network can be as shown in Figure 6 ; in the structure example diagram shown in Figure 6 , at each time step t, the outputs of the forward network and the backward network are calculated simultaneously (H and H ), the outputs of the two are connected and weighted to obtain the output y t of the BiLSTM. Wherein, the calculation formulas of the forward network and the backward network can be as shown in the following formula (4):

[0083]

[0084] Wherein, alpha represents a forward network output calculation weight, beta represents a backward network output calculation weight, and b y represents an output bias.

[0085] In the following, the flight trajectory prediction method of the aircraft shown in Figures 2-6 will be further explained and described in Figure 1 . Specifically:

[0086] In step S110, a motion equation of a hypersonic aircraft is established, and an original trajectory database of the hypersonic aircraft under different initial states and different flight control laws is determined based on the motion equation.

[0087] In the example embodiment, first, a motion equation of a hypersonic vehicle is established; specifically, this can be achieved by the following manner: obtaining a flight speed, a ballistic inclination angle, a ballistic deflection angle, a distance from the hypersonic vehicle to the center of the earth, a longitude, and a latitude of the hypersonic vehicle; in a ballistic coordinate system, a three-degree-of-freedom motion equation of the hypersonic vehicle is established according to the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic vehicle to the center of the earth, the longitude, and the latitude. Second, an original trajectory database of the hypersonic vehicle under different initial state variables and different flight control laws is determined based on the motion equation; specifically, this can be achieved by the following manner: first, different initial state variables of the hypersonic vehicle are determined based on the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic vehicle to the center of the earth, the longitude, and the latitude; second, different flight control laws of the hypersonic vehicle are determined according to an angle of attack control instruction and a roll angle control instruction of the hypersonic vehicle; then, the different initial state variables and the different flight control laws are input into the three-degree-of-freedom motion equation to obtain original flight trajectories under the different initial state variables and the different flight control laws; finally, the original trajectory database is generated according to the original flight trajectories and the initial state variables and the control laws corresponding to the original flight trajectories.

[0088] In the following, the example embodiment will be further explained with reference to the accompanying drawings. Figure 6 The specific establishment process of the original trajectory database will be further explained and described. Specifically, in the actual application process, considering various forces acting on the vehicle, a three-degree-of-freedom motion equation of the hypersonic vehicle is established in the ballistic coordinate system, which can be specifically shown in the following formula (5):

[0089]

[0090] wherein V is the speed of the vehicle, θ is the ballistic inclination angle, δ is the ballistic deflection angle, r is the distance from the vehicle to the center of the earth, λ is the longitude, and φ is the latitude; are differentials of V, θ, δ, r, λ, and φ, respectively;

[0091] μ M is the gravitational constant of the earth, ω e is the rotation speed of the earth, m is the mass of the hypersonic vehicle, υ is the roll angle instruction, which can be used to represent the included angle between the lift direction and the vertical plane containing the speed vector, s is the equivalent area of the vehicle, and ρ is the atmospheric density; X, Y, and Z are the drag, the lift, and the lateral force of the hypersonic vehicle, respectively, and the calculation formulae thereof can be shown in the following formula (6):

[0092]

[0093] Further, in the actual application process, if the lateral slip of the aircraft is not considered, β = 0; therefore the lateral force Z = 0; q represents the dynamic pressure constraint, and the specific calculation formula can be shown in the following formula (11); C L represents the lift coefficient, C D represents the drag coefficient, represents the lateral force coefficient, the lift coefficient, the drag coefficient and the lateral force coefficient can be obtained by inputting the current time angle of attack command and flight speed, and querying the interpolation table; further, the angle of attack-speed profile is set, the flight state of the aircraft is divided into two sections of initial dive section and descent section, and the function of the angle of attack changing with the speed can be shown in the following formula (7):

[0094]

[0095] The roll angle command is shown in the following formula:

[0096]

[0097] It can be seen that the state variables of the system are: V, θ, δ, r, λ, φ, the control input is the angle of attack command α and the roll angle command υ; the initial value of the state variable is set, the angle of attack command and the roll angle command are input into formula (5), and the system is integrated, so that the flight trajectory of the aircraft can be obtained. At the same time, the hypersonic aircraft needs to meet the heat flux constraint, the overload constraint and the dynamic pressure constraint in the flight process. Therefore, the heat flux density constraint, the overload constraint and the dynamic pressure constraint need to be established. Among them, the heat flux density constraint can be shown in the following formula (9):

[0098]

[0099] Wherein, Q represents the heat flux density, represents the differential of the heat flux density, k Q is a constant coefficient related to the shape and shape at the stagnation point, k Q = 6 x 10 -8 , v is the speed of the aircraft, and p is the atmospheric density, which is represented in a simplified exponential form as shown in the following formula (10):

[0100]

[0101] Wherein, p0 = 1.225 g / L, h s = 7200 m; H is the flight height of the hypersonic aircraft.

[0102] Further, the dynamic pressure constraint can be shown in the following formula (11):

[0103]

[0104] Further, the overload constraint can be shown as formula (12) as follows:

[0105]

[0106] wherein L and D represent lift and drag of the aircraft respectively, a is the angle of attack, m is the mass of the aircraft, g0 is the acceleration of gravity at sea level, n ymax represents the maximum overload value allowed by the aircraft.

[0107] Further, in the actual application process, the flight state of the hypersonic aircraft can be analyzed from the longitudinal and lateral planes; and when the longitudinal plane is in force balance, the aircraft is in balanced glide state; otherwise, the aircraft is in jump glide state; meanwhile, since the trajectory of the balanced glide state is relatively flat, the trajectory prediction is relatively low in difficulty, therefore, the flight trajectory prediction method of the aircraft recorded in the example embodiments of the present disclosure focuses on the trajectory prediction of the jump glide aircraft. Further, in the process of trajectory iteration calculation, the main factors affecting the trajectory longitudinal plane are the initial speed and height of the aircraft and the aerodynamic lift generated by the angle of attack of the aircraft, and the main factors affecting the lateral plane motion of the aircraft are the initial speed and height of the aircraft and the aerodynamic lateral turning force generated by the roll angle of the aircraft. Therefore, the construction of the trajectory database needs to give the initial flight conditions and the angle of attack control instruction and the roll angle control instruction. In the actual application process, as shown in formula (13), the Monte Carlo trajectory database simulation method is implemented as follows: Figure 7

[0108] Step 1: Given the value range of the independent variable:

[0109] 0≤υ≤υ max , 20km≤h≤90km, 0≤α≤α max , 2000m / s≤V≤V max .

[0110] Step 2.1: Start the loop, let i=0, j=0, k=0; and initialize the given independent variable parameters respectively: υ i =υ max / n×i, h i =(h max -h min ) / n×i, α0=α max / n×i; wherein n is a positive integer, representing the degree of trajectory division; the roll angle control instruction adopts a constant instruction, and the expression formula is as follows: υ(t)=υ i ;

[0111] ​Step 2.2: Substitute the roll angle instruction formula and the angle of attack instruction calculation formula into the aircraft trajectory calculation formula for integral iteration, and record the longitude, latitude and height data of the simulation trajectory;

[0112] Step 2.3: i = i + 1, j = j + 1, k = k + 1, respectively change the initial values of velocity, height and angle of attack, and perform trajectory calculation;

[0113] Step 3: When i≤n, cycle Step 2.1-Step 2.3, and record input υ i , α i , v i , h i and output λ i , φ i , h i .

[0114] It needs to be supplemented here that the reason for generating the original trajectory database is that in the trajectory prediction problem, since the target is often non-cooperative, it is difficult to accurately judge the flight law of the target in a short time. In order to solve this problem, a trajectory database of the gliding projectile (i.e. the original trajectory database) is obtained based on Monte Carlo for different initial states and flight control laws of the aircraft, and then the simulation trajectory data is pre-trained 1DCNN-BiLSTM network to complete the pre-training of multiple neural network models.

[0115] In step S120, the network model to be trained is trained based on the original trajectory database to obtain a current trajectory prediction model.

[0116] Specifically, the network model to be trained is trained based on the original trajectory database to obtain a current trajectory prediction model, which can be realized by the following manner: inputting the initial state variables and control laws corresponding to the original flight trajectory in the original trajectory database into the network model to be trained to obtain a current prediction result; constructing a current loss function according to the current prediction result and the original flight trajectory, and adjusting the parameters in the network model to be trained based on the current loss function to obtain a current trajectory prediction model. That is, in the process of actual application, the original flight trajectory can be calculated based on the above-mentioned formula (5), and the parameters used to calculate the original flight trajectory can be input into the network model to be trained to obtain the current prediction result; at the same time, when constructing the current loss function, a mean square error loss function can be used, or a cross-entropy loss function can be used, and the present example does not specially limit this.

[0117] In step S130, a trajectory prediction model to be migrated is determined from the current trajectory prediction model according to the target flight trajectory of the aircraft to be predicted and the original flight trajectories in the original trajectory database, and model parameters in the trajectory prediction model to be migrated are migrated to the trajectory prediction model to be corrected.

[0118] In the example embodiment, first, a trajectory prediction model to be migrated is determined from the current trajectory prediction model according to the target flight trajectory of the aircraft to be predicted and the original flight trajectories in the original trajectory database. Specifically, the following method can be used: first, the target flight trajectory of the aircraft to be predicted is obtained, and noise reduction processing is performed on the target flight trajectory to obtain a flight trajectory after noise reduction processing; second, a first trajectory sequence of the flight trajectory after noise reduction processing is determined, and a second trajectory sequence of the original flight trajectories in the original trajectory database is determined; then, the similarity of the first trajectory sequence and the second trajectory sequence is calculated, and the trajectory prediction model to be migrated is determined from the current trajectory prediction model according to the similarity.

[0119] In an example embodiment, the similarity of the first trajectory sequence and the second trajectory sequence is calculated, and the trajectory prediction model to be migrated is determined from the current trajectory prediction model according to the similarity, which can be implemented in the following manner: the Euclidean distance between the first trajectory sequence and the second trajectory sequence is calculated, and the similarity of the first trajectory sequence and the second trajectory sequence is determined according to the Euclidean distance; the original flight trajectories are sorted based on the similarity of the first trajectory sequence and the second trajectory sequence, and the original flight trajectory with the smallest similarity is extracted from the original flight trajectories based on the trajectory sorting result; the current trajectory prediction model corresponding to the original flight trajectory with the smallest similarity is taken as the trajectory prediction model to be migrated.

[0120] In the following, the specific determination process of the trajectory prediction model to be migrated will be further explained and described. Specifically, in actual application, if the flight trajectory of the aircraft to be predicted needs to be predicted, the information of the aircraft to be predicted needs to be observed based on a monitoring system; at the same time, for the observed unknown information of the aircraft, since the observation system has certain noise, data needs to be cleaned of outliers and filled with missing values, and position information needs to be filtered and smoothed. For example, assuming that the mean μ and standard deviation σ of each adjacent three-point gliding trajectory are calculated, in the noise reduction processing, data exceeding (μ-3σ, μ+3σ) can be removed as outliers, and then the missing data after removing outliers is calculated and filled with the data of the adjacent two points. The formula for supplementing is shown in the following formula (13):

[0121] y(d, t) = (y(d, t - 1) + y(d, t + 1)) / 2; equation (13)

[0122] Wherein, y(d, t) represents the vacancy value supplement, which is obtained by linear average of data at previous and next time.

[0123] Further, after obtaining the flight trajectory after noise reduction processing, the similarity of the first trajectory sequence and the second trajectory sequence can be calculated; wherein, in the process of similarity selection, the position information history sequence of the target aircraft can be respectively calculated with the trajectory information in the trajectory library at the same time step, or the similarity of the target flight trajectory and each trajectory in the trajectory library; wherein, the calculation of trajectory similarity can use the spatial Euclidean distance of the target trajectory and the database trajectory to represent; wherein, the specific calculation process of the Euclidean distance can be shown in the following formula (14):

[0124]

[0125] Wherein, d(x, y, z) is the Euclidean distance, x r , y r , z r is the first trajectory sequence, x p , y p , z p is the second trajectory sequence. Further, after obtaining the Euclidean distance, the similarity of the target trajectory and the database can be obtained, and the smaller the Euclidean distance is, the smaller the trajectory gap is and the higher the similarity is. Select the source domain data with the smallest trajectory similarity as the migration data (the trajectory prediction model to be migrated); wherein, the specific selection process of the trajectory prediction model to be migrated can be shown in the following formula (15):

[0126]

[0127] Secondly, after obtaining the trajectory prediction model to be migrated, the model parameters in the trajectory prediction model to be migrated can be migrated to the trajectory prediction model to be corrected. Specifically, the migration of the parameters can be achieved by extracting the model parameters in the trajectory prediction model to be migrated and performing parameter initialization processing on the model parameters in the trajectory prediction model to be corrected based on the extracted model parameters. Specifically, the migration process described herein can also be referred to as transfer learning. Transfer learning is a branch of machine learning, and mainly involves two domains, a source domain and a target domain. The source of the migration is referred to as the source domain (Ds), and the destination of the migration information is referred to as the target domain (Dt). Transfer learning needs to use the model parameters learned in the source domain to perform knowledge migration on the small sample unknown model in the target domain. In this example embodiment, the source domain is the original trajectory database described above, and the target domain is the target flight trajectory of the aircraft to be predicted. Further, in actual application, transfer learning first trains all the trajectory information in the source domain and saves the trained network model. Then, the network model parameters in the source domain are migrated to the target domain, and the network parameters in the source domain that are most similar to the target domain dataset are selected for fine-tuning. Finally, the target trajectory feature values obtained by the CNN network are fixed, and the BiLSTM network parameters are fine-tuned. The adjusted network model is used as a prediction model for prediction of the target domain data.

[0128] In step S140, the trajectory prediction model to be corrected is parameter-corrected based on the target flight trajectory, to obtain a target trajectory prediction model, and the flight trajectory of the aircraft to be predicted is predicted based on the target trajectory prediction model.

[0129] In this example embodiment, first, the trajectory prediction model to be corrected is parameter-corrected based on the target flight trajectory, to obtain a target trajectory prediction model. Specifically, the target prediction result can be obtained by first obtaining the sample data of the aircraft to be predicted and inputting the sample data into the trajectory prediction model to be corrected. Secondly, a target loss function is constructed based on the target prediction result and the target flight trajectory, and the trajectory prediction model to be corrected is parameter-corrected based on the target loss function, to obtain a target trajectory prediction model. The trajectory prediction model to be corrected can be parameter-corrected based on the target loss function, to obtain a target trajectory prediction model, by fixing the weight values of the convolutional neural network and fine-tuning the weight values and bias of the bidirectional long short-term memory network model based on the target loss function, to obtain a target trajectory prediction model. In the process of fine-tuning the weight values and bias of the bidirectional long short-term memory network model, the selected optimizer includes an Adam optimizer that adjusts the learning rate adaptively.

[0130] Hereinafter, the specific determination process of the target trajectory prediction model will be further explained and described. Specifically, in the actual application process, after the 1DCNN-BiLSTM network model parameters with the highest similarity are migrated to the target domain prediction model as initial parameters, the weights of the CNN network can be fixed, and the Bi-LSTM network weights and biases can be fine-tuned using the observed sample data; at the same time, in the parameter fine-tuning process, the optimizer selects the Adam optimizer which can adaptively adjust the learning rate, and the initial learning rate is adjusted to 0.005, and its calculation formula can be shown in the following formula (16):

[0131]

[0132] where g t represents the gradient of the t-th time step, represents the square of the gradient of the t-th time step, and the optimizer can accumulate the historical data of each gradient parameter, and adaptively adjust the learning rate as the number of learning increases. In the parameter fine-tuning process, the network fine-tuning parameter settings can be shown in Table 1 as follows:

[0133] Table 1 Network fine-tuning parameter settings

[0134]

[0135] Further, in order to improve the speed of model migration and fine-tuning, the number of nodes of the LSTM unit needs to be controlled, and the number of hidden nodes is defined as 64. The specific settings of the internal unit structure of the LSTM can be shown in Table 2 as follows:

[0136] Table 2 LSTM network structure

[0137]

[0138] Finally, after obtaining the target trajectory prediction model, the target trajectory prediction model needs to be evaluated. Specifically, in the model evaluation, the test data can be input into the trained prediction model, and the root mean square error function (RMSE) is used as the accuracy evaluation index to evaluate the model. Specifically, the root mean square error function can be shown in the following formula (17):

[0139]

[0140] where M is the number of data samples in the network prediction set; is the de-normalized network prediction value; y i is the test data label value.

[0141] So far, the specific fine-tuning process of the target trajectory prediction model has been completed. Further, after obtaining the target trajectory prediction model, the flight trajectory of the to-be-predicted aircraft can be predicted based on the target trajectory prediction model. Specifically, the flight trajectory of the to-be-predicted aircraft can be predicted by the following method: obtaining the flight speed, ballistic inclination angle, ballistic declination angle, distance from the hypersonic aircraft to the center of the earth, longitude, and latitude of the to-be-predicted aircraft; constructing a model input vector according to the flight speed, ballistic inclination angle, ballistic declination angle, distance from the hypersonic aircraft to the center of the earth, longitude, and latitude of the to-be-predicted aircraft; inputting the model input vector into the target trajectory prediction model to obtain the flight trajectory prediction result of the to-be-predicted aircraft; and the specific prediction process can be referred to Figure 8

[0142] It should be noted here that the reason why the target trajectory prediction model needs to be determined and the flight trajectory of the to-be-predicted aircraft is predicted based on the target trajectory prediction model is that when the to-be-predicted aircraft is observed, the mass and the equivalent area of the to-be-predicted aircraft cannot be accurately known, so the flight trajectory of the to-be-predicted aircraft cannot be accurately predicted based on the formula (5) described above. The present application determines the target trajectory prediction model and predicts the flight trajectory of the to-be-predicted aircraft based on the target trajectory prediction model, thereby solving the problem.

[0143] The following is an apparatus embodiment of the present disclosure, which can be used to execute the method embodiments of the present disclosure. For details not disclosed in the apparatus embodiment of the present disclosure, please refer to the method embodiments of the present disclosure.

[0144] The example embodiments of the present disclosure also provide an aircraft flight trajectory prediction apparatus. Specifically, referring to FIG. 9, the aircraft flight trajectory prediction apparatus can include an original trajectory database construction module 910, a network model training module 920, a model parameter migration module 930, and a flight trajectory prediction module 940. Wherein: Figure 9 The original trajectory database construction module 910 can be used to establish the motion equation of the hypersonic aircraft, and determine the original trajectory database of the hypersonic aircraft under different initial states and different flight control laws based on the motion equation;

[0145] The network model training module 920 can be used to train the to-be-trained network model based on the original trajectory database to obtain a current trajectory prediction model;

[0146]

[0147] ​​The model parameter migration module 930 can be configured to determine a trajectory prediction model to be migrated from the current trajectory prediction model according to a target flight trajectory of an aircraft to be predicted and original flight trajectories in an original trajectory database, and migrate model parameters in the trajectory prediction model to be migrated to a trajectory prediction model to be corrected.

[0148] The flight trajectory prediction module 940 can be configured to perform parameter correction on the trajectory prediction model to be corrected based on the target flight trajectory, to obtain a target trajectory prediction model, and predict a flight trajectory of the aircraft to be predicted based on the target trajectory prediction model.

[0149] In an example embodiment of the present disclosure, a motion equation of a hypersonic aircraft is established, including: obtaining a flight speed, a ballistic inclination angle, a ballistic deflection angle, a distance from the hypersonic aircraft to the center of the earth, a longitude, and a latitude of the hypersonic aircraft; and in a ballistic coordinate system, establishing a three-degree-of-freedom motion equation of the hypersonic aircraft according to the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude, and the latitude.

[0150] In an example embodiment of the present disclosure, an original trajectory database of a hypersonic aircraft under different initial states and different flight control laws is determined based on a motion equation, including: determining different initial state variables of the hypersonic aircraft based on a flight speed, a ballistic inclination angle, a ballistic deflection angle, a distance from the hypersonic aircraft to the center of the earth, a longitude, and a latitude; determining different flight control laws of the hypersonic aircraft according to an angle of attack control instruction and a roll angle control instruction; inputting the different initial state variables and the different flight control laws into the three-degree-of-freedom motion equation to obtain original flight trajectories under the different initial state variables and the different flight control laws; and generating the original trajectory database according to the original flight trajectories and initial state variables and control laws corresponding to the original flight trajectories.

[0151] In an example embodiment of the present disclosure, a network model to be trained is trained based on the original trajectory database to obtain a current trajectory prediction model, including: inputting initial state variables and control laws corresponding to original flight trajectories in the original trajectory database into the network model to be trained to obtain a current prediction result; constructing a current loss function according to the current prediction result and the original flight trajectories, and adjusting parameters in the network model to be trained based on the current loss function to obtain the current trajectory prediction model.

[0152] In an example embodiment of the present disclosure, the trajectory prediction model to be migrated is determined from the current trajectory prediction model according to a target flight trajectory of a to-be-predicted aircraft and original flight trajectories in an original ballistic database, comprising: obtaining the target flight trajectory of the to-be-predicted aircraft, and performing noise reduction processing on the target flight trajectory to obtain a flight trajectory after noise reduction processing; determining a first trajectory sequence of the flight trajectory after noise reduction processing, and determining a second trajectory sequence of the original flight trajectories in the original ballistic database; calculating the similarity of the first trajectory sequence and the second trajectory sequence, and determining the trajectory prediction model to be migrated from the current trajectory prediction model according to the similarity.

[0153] In an example embodiment of the present disclosure, the similarity of the first trajectory sequence and the second trajectory sequence is calculated, and the trajectory prediction model to be migrated is determined from the current trajectory prediction model according to the similarity, comprising: calculating the Euclidean distance between the first trajectory sequence and the second trajectory sequence, and determining the similarity of the first trajectory sequence and the second trajectory sequence according to the Euclidean distance; sorting the original flight trajectories based on the similarity of the first trajectory sequence and the second trajectory sequence, and extracting the original flight trajectory with the smallest similarity from the original flight trajectories based on the trajectory sorting result; taking the current trajectory prediction model corresponding to the original flight trajectory with the smallest similarity as the trajectory prediction model to be migrated.

[0154] In an example embodiment of the present disclosure, the model parameters in the trajectory prediction model to be migrated are migrated to the trajectory prediction model to be corrected, comprising: extracting the model parameters in the trajectory prediction model to be migrated, and performing parameter initialization processing on the model parameters in the trajectory prediction model to be corrected based on the extracted model parameters to realize the migration processing of the parameters.

[0155] In an example embodiment of the present disclosure, the trajectory prediction model to be corrected is parameter corrected based on the target flight trajectory to obtain a target trajectory prediction model, comprising: obtaining sample data of the to-be-predicted aircraft observed, and inputting the sample data into the trajectory prediction model to be corrected to obtain a target prediction result; constructing a target loss function according to the target prediction result and the target flight trajectory, and performing parameter correction on the trajectory prediction model to be corrected based on the target loss function to obtain a target trajectory prediction model.

[0156] In an example embodiment of the present disclosure, the trajectory prediction model to be corrected comprises a convolutional neural network model and a bidirectional long short-term memory network model; wherein the parameter correction of the trajectory prediction model to be corrected based on the target loss function obtains a target trajectory prediction model, comprising: fixing the weight value of the convolutional neural network, and fine-tuning the weight value and the bias of the bidirectional long short-term memory network model based on the target loss function to obtain the target trajectory prediction model; wherein in the process of fine-tuning the weight value and the bias of the bidirectional long short-term memory network model, the selected optimizer comprises an Adam optimizer with adaptive adjustment of learning rate.

[0157] In an example embodiment of the present disclosure, the prediction of the flight trajectory of the to-be-predicted aircraft based on the target trajectory prediction model comprises: obtaining the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude and the latitude of the to-be-predicted aircraft; constructing a model input vector according to the flight speed, the ballistic inclination angle, the ballistic deflection angle, the distance from the hypersonic aircraft to the center of the earth, the longitude and the latitude of the to-be-predicted aircraft; inputting the model input vector into the target trajectory prediction model to obtain the flight trajectory prediction result of the to-be-predicted aircraft.

[0158] The specific details of each module in the flight trajectory prediction device of the aircraft have been described in detail in the corresponding flight trajectory prediction method of the aircraft, so this will not be described here.

[0159] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, such division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into several modules or units embodied.

[0160] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this is not required or implied that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. In addition or alternatively, some steps can be omitted, a plurality of steps can be combined into one step, and / or one step can be divided into a plurality of steps, etc.

[0161] In an example embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0162] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0163] The following reference Figure 10 To describe an electronic device 1000 according to such an embodiment of the present disclosure. Figure 10 The electronic device 1000 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0164] like Figure 10 As shown, the electronic device 1000 is manifested in the form of a general-purpose computing device. The components of the electronic device 1000 may include, but are not limited to: at least one processing unit 1010, at least one storage unit 1020, a bus 1030 connecting different system components (including storage unit 1020 and processing unit 1010), and a display unit 1040.

[0165] The storage unit stores program code that can be executed by the processing unit 1010, causing the processing unit 1010 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 1010 can perform actions such as... Figure 1 The steps shown are as follows: Step S110: Establish the motion equations of the hypersonic vehicle, and determine the original ballistic database of the hypersonic vehicle under different initial states and different flight control laws based on the motion equations; Step S120: Train the network model to be trained based on the original ballistic database to obtain the current trajectory prediction model; Step S130: Determine the trajectory prediction model to be transferred from the current trajectory prediction model according to the target flight trajectory of the vehicle to be predicted and the original flight trajectory in the original ballistic database, and transfer the model parameters in the trajectory prediction model to be transferred to the trajectory prediction model to be corrected; Step S140: Correct the parameters of the trajectory prediction model to be corrected based on the target flight trajectory to obtain the target trajectory prediction model, and predict the flight trajectory of the vehicle to be predicted based on the target trajectory prediction model.

[0166] Storage unit 1020 may include readable media in the form of volatile storage units, such as random access memory (RAM) 10201 and / or cache memory 10202, and may further include read-only memory (ROM) 10203.

[0167] The storage unit 1020 can also include a program / utility 10204 having a set of programs / modules 10205, each of which performs one or more tasks. The programs and modules 10205 can include, but are not limited to, one or more of: an operating system, one or more application programs, other program modules, and program data, each of which can include implementations of the network environment, alone or in combination.

[0168] The bus 1030 can represent one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics bus, a processor or local bus using any of a variety of bus architectures.

[0169] The electronic device 1000 can also communicate with one or more external devices 1100 such as a keyboard or pointing device, a Bluetooth device, etc.; other devices such as a storage device or an external effects device; and / or one or more devices that enable a user to interact with the electronic device 1000 in some way (e.g., a sensor, etc.). Such interaction can involve the user's input or responses to various aspects of the electronic device 1000. Further, the electronic device 1000 can communicate with one or more devices that enable the electronic device 1000 to perform a function. The communication can be via the input / output (I / O) interface 1050. Still yet, the electronic device 1000 can communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or the Internet) through a network adapter 1060. It will be appreciated that the network adapter 1060 can be communicatively coupled to the other

[0170] Those skilled in the art will readily understand that the example embodiments described herein can be implemented by software and / or by software in combination with the necessary hardware. Thus, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, or the like) or a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to perform the methods according to the embodiments of the present disclosure.

[0171] In exemplary embodiments of the present disclosure, a computer readable storage medium having stored thereon a program product capable of implementing the above-described methods of the present specification is also provided. In some possible implementations, various aspects of the present disclosure can also be implemented in the form of a program product including a program code that, when run on a terminal device, causes the terminal device to perform the steps described in the above "Exemplary Methods" section according to various exemplary embodiments of the present disclosure.

[0172] A program product for implementing the above-described methods according to embodiments of the present disclosure can take the form of a portable compact disc read-only memory (CD-ROM) and include a program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in the present document, a readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0173] The program product can take any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium, for example, can be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0174] The computer readable signal medium can include a data signal propagated in baseband or propagated as a carrier wave in a propagated data signal, in which the readable program code is embodied. Such propagated data signal can take multiple forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. The readable signal medium can also be any readable medium that is not a readable storage medium and that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0175] The program code contained on the readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, and the like, or any suitable combination thereof.

[0176] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0177] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0178] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not invented by this disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A method for predicting the flight trajectory of an aircraft, characterized in that, include: Establish the equations of motion for a hypersonic vehicle, and based on these equations, determine the original ballistic database of the hypersonic vehicle under different initial states and different flight control laws. The network model to be trained is trained based on the original ballistic database to obtain the current trajectory prediction model; Based on the target flight trajectory of the aircraft to be predicted and the original flight trajectory in the original ballistic database, the trajectory prediction model to be transferred is determined from the current trajectory prediction model, and the model parameters in the trajectory prediction model to be transferred are transferred to the trajectory prediction model to be corrected. Based on the target flight trajectory, the parameters of the trajectory prediction model to be corrected are corrected to obtain the target trajectory prediction model, and the flight trajectory of the aircraft to be predicted is predicted based on the target trajectory prediction model; wherein, the trajectory prediction model to be corrected includes a convolutional neural network model and a bidirectional long short-term memory network model. The trajectory prediction model to be transferred is determined as follows: the Euclidean distance between the first trajectory sequence corresponding to the target flight trajectory and the second trajectory sequence corresponding to the original flight trajectory is calculated, and the similarity between the first trajectory sequence and the second trajectory sequence is determined based on the Euclidean distance; the original flight trajectory is sorted based on the similarity between the first trajectory sequence and the second trajectory sequence, and the original flight trajectory with the smallest similarity is extracted from the original flight trajectory based on the trajectory sorting result; the current trajectory prediction model corresponding to the original flight trajectory with the smallest similarity is used as the trajectory prediction model to be transferred.

2. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Establish the equations of motion for a hypersonic vehicle, including: To obtain the flight speed, trajectory inclination angle, trajectory deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude of the hypersonic vehicle; In the ballistic coordinate system, based on the flight speed, ballistic inclination angle, ballistic deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude, the three-degree-of-freedom equations of motion for the hypersonic vehicle are established.

3. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Based on the equations of motion, a raw ballistic database of hypersonic vehicles under different initial states and different flight control laws is determined, including: The different initial state variables of the hypersonic vehicle are determined based on flight speed, ballistic inclination angle, ballistic deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude. Based on the angle-of-attack control command and the roll angle control command of the hypersonic vehicle, the different flight control laws of the hypersonic vehicle are determined; By inputting the different initial state variables and different flight control laws into the three-degree-of-freedom equations of motion, the original flight trajectories under different initial state variables and different flight control laws are obtained; The original ballistic database is generated based on the original flight trajectory and the corresponding initial state variables and control laws.

4. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, The network model to be trained is trained based on the original ballistic database to obtain the current trajectory prediction model, including: The initial state variables and control laws corresponding to the original flight trajectory in the original ballistic database are input into the network model to be trained to obtain the current prediction result; Based on the current prediction result and the original flight trajectory, a current loss function is constructed, and the parameters in the network model to be trained are adjusted based on the current loss function to obtain the current trajectory prediction model.

5. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Based on the target flight trajectory of the aircraft to be predicted and the original flight trajectories in the original ballistic database, determine the trajectory prediction model to be transferred from the current trajectory prediction model, including: The target flight trajectory of the aircraft to be predicted is obtained, and the target flight trajectory is denoised to obtain the denoised flight trajectory. Determine the first trajectory sequence of the flight trajectory after noise reduction processing, and determine the second trajectory sequence of the original flight trajectory in the original ballistic database; Calculate the similarity between the first trajectory sequence and the second trajectory sequence, and determine the trajectory prediction model to be transferred from the current trajectory prediction model based on the similarity.

6. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Transferring model parameters from the trajectory prediction model to the trajectory prediction model to be corrected includes: The model parameters in the trajectory prediction model to be transferred are extracted, and the model parameters in the trajectory prediction model to be corrected are initialized based on the extracted model parameters to achieve parameter transfer processing.

7. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Based on the target flight trajectory, the trajectory prediction model to be corrected is parameter-corrected to obtain the target trajectory prediction model, including: Acquire sample data of the observed aircraft to be predicted, and input the sample data into the trajectory prediction model to be corrected to obtain the target prediction result; Based on the target prediction results and the target flight trajectory, a target loss function is constructed, and the parameters of the trajectory prediction model to be corrected are corrected based on the target loss function to obtain the target trajectory prediction model.

8. The flight trajectory prediction method for an aircraft according to claim 7, characterized in that, The trajectory prediction model to be corrected is parameter-corrected based on the target loss function to obtain the target trajectory prediction model, including: The weights of the convolutional neural network are fixed, and the weights and biases of the bidirectional long short-term memory network model are fine-tuned based on the target loss function to obtain the target trajectory prediction model. In the process of fine-tuning the weights and biases of the bidirectional long short-term memory network model, the selected optimizer includes the Adam optimizer with adaptive learning rate adjustment.

9. The flight trajectory prediction method for an aircraft according to claim 1, characterized in that, Based on the target trajectory prediction model, the flight trajectory of the aircraft to be predicted is predicted, including: The flight speed, trajectory inclination angle, trajectory deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude of the aircraft to be predicted are obtained. The model input vector is constructed based on the flight speed, trajectory tilt angle, trajectory deflection angle, distance from the hypersonic vehicle to the Earth's center, longitude, and latitude of the aircraft to be predicted. The model input vector is input into the target trajectory prediction model to obtain the flight trajectory prediction result of the aircraft to be predicted.

Images