CN-121457522-B - Method for predicting motion trail of cross-medium aircraft and computer program product

CN121457522BCN 121457522 BCN121457522 BCN 121457522BCN-121457522-B

Abstract

The application discloses a motion trail prediction method of a cross-medium aircraft and a computer program product. The method comprises the steps of obtaining observation state parameters of a medium-crossing aircraft at a current time point, inputting the observation state parameters of the current time point into a trained track prediction model and outputting prediction state parameters of a next time point, wherein the trained track prediction model comprises a trained state prediction model and a trained cyclic neural network, the trained state prediction model is used for merging the prediction state parameters of the previous time point and the observation state parameters of the current time point to obtain corrected prediction state parameters of the current time point, calculating the prediction state of the next time point according to the corrected prediction state parameters of the current time point, and the trained cyclic neural network is used for calculating weight coefficients during merging and determining a motion track of the medium-crossing aircraft according to the prediction state parameters of the medium-crossing aircraft at the next time point. The method and the device can improve the accuracy of the motion trail prediction of the cross-medium aircraft.

Inventors

CHEN YINGLIANG
LI HONGYUAN
LUO YIJIA
LIU JIALIN
Zhu Jinzhuo
DUAN HUILING

Assignees

北京大学
中国船舶重工集团公司第七〇五研究所

Dates

Publication Date: 20260508
Application Date: 20260105

Claims (10)

1. The motion trail prediction method of the cross-medium aircraft is characterized by comprising the following steps of: Acquiring observation state parameters of the cross-medium aircraft at the current time point; Inputting the observation state parameters of the current time point into a trained track prediction model, outputting the prediction state parameters of the cross-medium aircraft at the next time point by adopting the trained track prediction model, wherein the trained track prediction model comprises a trained state prediction model and a trained cyclic neural network, the trained state prediction model comprises the prediction state parameters of the cross-medium aircraft at the previous time point, the trained state prediction model is used for fusing the prediction state parameters of the previous time point and the observation state parameters of the current time point to obtain corrected prediction state parameters of the current time point, calculating the prediction state of the next time point according to the corrected prediction state parameters of the current time point, and the trained cyclic neural network is used for calculating a weight coefficient between the prediction state parameters of the previous time point and the observation state parameters of the current time point during fusion; And determining the motion trail of the medium-crossing aircraft according to the predicted state parameters of the medium-crossing aircraft at the next time point.
2. The method of claim 1, wherein the determining the movement trajectory of the cross-medium vehicle based on the predicted state parameters of the cross-medium vehicle at the next point in time comprises: Analyzing the corrected prediction state parameters of the current time point to obtain corrected position information of the cross-medium aircraft at the current time point; analyzing the predicted state parameter of the next time point to obtain the predicted position information of the cross-medium aircraft at the next time point; And determining the motion trail according to the corrected position information and the predicted position information.
3. The method of claim 1, wherein the trained state prediction model further comprises observed predicted state parameters of the cross-medium vehicle at a current point in time and observed state parameters at a previous point in time, the outputting the predicted state parameters of the cross-medium vehicle at the next point in time using the trained trajectory prediction model comprising: Calculating at least one of an observed difference value of a current time point, new updated information of the current time point, a forward transition difference value of the current time point and a forward update difference value of the current time point, wherein the observed difference value of the current time point is a difference value of an observed state parameter of the previous time point and an observed state parameter of the current time point, the new updated information of the current time point is a difference value of an observed state parameter of the current time point and an observed prediction state parameter of the current time point, the forward transition difference value of the current time point is a difference value of a prediction state parameter of the current time point and a prediction state parameter of the previous time point, and the forward update difference value of the current time point is a difference value of a corrected prediction state parameter of the current time and a prediction state parameter of the current time point; inputting at least one of the observed difference value of the current time point, the new updated information of the current time point, the forward transfer difference value of the current time point and the forward updated difference value of the current time point into the trained cyclic neural network, and outputting the weight coefficient by adopting the trained cyclic neural network; And fusing the predicted state parameters of the previous time point and the observed state parameters of the current time point based on the weight coefficient by adopting the trained track prediction model to obtain corrected predicted state parameters of the current time point, and calculating and outputting predicted state parameters of the next time point according to the corrected predicted state parameters of the current time point.
4. A method according to claim 3, wherein the trained trajectory prediction model has a trained state transfer function, and the calculating and outputting the predicted state parameters for the next time point from the corrected predicted state parameters for the current time point comprises: And inputting the corrected predicted state parameters of the current time point into the trained state transfer function, calculating the predicted state parameters of the next time point by adopting the trained state transfer function, and outputting the predicted state parameters.
5. The method of claim 1, wherein the trained trajectory prediction model has a trained observation function, and wherein after the entering of the observation state parameters for the current point in time into the trained trajectory prediction model, further comprises: Inputting the corrected predicted state parameters of the current time point into the trained observation function, calculating the predicted observed state parameters of the next time point by adopting the trained observation function, and outputting the predicted observed state parameters; acquiring an observation state parameter of the cross-medium aircraft at the next time point; Calculating a loss value of the trained track prediction model at the current time point according to the predicted observation parameter of the next time point and the observation state parameter of the next time point; and updating parameters of the trained track prediction model according to the loss value of the current time point.
6. The method of claim 5, wherein updating parameters of the trained trajectory prediction model based on the loss value at the current point in time comprises: Updating parameters of the trained cyclic neural network by adopting a gradient descent method; and updating the state transition coefficient of the trained state prediction model by adopting a gradient descent method.
7. The method of claim 1, wherein before the inputting the observed state parameter for the current point in time into the trained trajectory prediction model, further comprises: acquiring training data, wherein the training data comprises observation state parameters of the cross-medium aircraft at a plurality of continuous historical time points; Acquiring a preset track prediction model, wherein the preset track prediction model comprises a preset state prediction model and a preset cyclic neural network, and the preset state prediction model comprises a preset state transfer function and a preset observation function; And training the preset track prediction model by adopting the training data until the loss value of the preset track prediction model reaches a preset threshold value to obtain a trained track prediction model.
8. The method of claim 7, wherein training the pre-set trajectory prediction model using the training data until a loss value of the pre-set trajectory prediction model reaches a pre-set threshold value, comprises: inputting a continuous observation state parameter sequence in the training data into the preset track prediction model; calculating a predicted state parameter sequence and a predicted observed state parameter sequence corresponding to the observed state parameter sequence by adopting the preset state prediction model; calculating an observation difference value sequence, a new update information sequence, a forward transfer difference value sequence and a forward update difference value sequence according to the observation state parameter sequence, the prediction state parameter sequence and the prediction observation state parameter sequence; Calculating the variation posterior distribution of potential variables at the latest time point in an observation state parameter sequence according to the observation difference value sequence, the new update information sequence, the forward transfer difference value sequence and the forward update difference value sequence by adopting the preset cyclic neural network; sampling the variation posterior distribution to obtain a potential variable sample; Generating a reconstruction observation data distribution corresponding to the potential variable sample by adopting a preset observation function of the preset state prediction model; calculating a loss value of a preset track prediction model according to the reconstructed observation data distribution and the observation state parameters of the latest time point in the observation state parameter sequence; And updating parameters of the preset state prediction model and parameters of the preset cyclic neural network based on the loss value.
9. The method of claim 8, wherein calculating a loss value of a preset trajectory prediction model based on the reconstructed observed data distribution and the observed state parameters at the latest time point in the sequence of observed state parameters, comprises: Calculating the loss value based on a variant lower bound loss function, the variant lower bound loss function being: , Wherein, the For the lower variation bound, x is the sequence of observed state parameters, For the parameters of the preset trajectory prediction model, For the variational posterior parameters, z is a latent variable, The reconstruction loss term is expected to be a function of, For the variational posterior distribution, pi (z) is a priori, KL divergence; the updating the parameters of the preset state prediction model and the parameters of the preset cyclic neural network based on the loss value comprises the following steps: and maximizing the variation lower bound through a gradient descent method, and updating parameters of the preset state prediction model and parameters of the preset cyclic neural network.
10. A computer program product, characterized in that instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform the method of any of claims 1-9.

Description

Method for predicting motion trail of cross-medium aircraft and computer program product Technical Field The application belongs to the technical field of automatic control and navigation, and particularly relates to a motion trail prediction method of a cross-medium aircraft and a computer program product. Background The medium-crossing aircraft is novel intelligent equipment capable of freely switching among different mediums such as air, water and the like and navigating efficiently, and the air-water seamless transition capability of the intelligent equipment has important application value in the fields of military reconnaissance, ocean exploration, disaster rescue and the like. When crossing the water-air interface, the cross-medium aircraft can face multiple dynamics challenges such as abrupt change of flow field characteristics, rapid decrease of control stability, disturbance of complex environment and the like. In the process, the motion trail is predicted with high precision, which is a key technical premise for realizing autonomous navigation, risk avoidance and task optimization. Currently, a prediction method based on kalman filtering is a common technology for track prediction. However, the premise of such predictive methods is that accurate prediction of trajectories must be provided with accurate mathematical models describing how the vehicle moves and statistical properties of internal and external disturbances during movement of the vehicle. However, in the scene that the movement rule of the cross medium is changed drastically, neither a unified and accurate mathematical model can describe the whole navigation process of the cross medium, nor the statistical rule of various complex interferences can be known in advance. Therefore, the existing method for predicting the motion trail of the cross-medium aircraft has the problem of insufficient prediction precision. Disclosure of Invention The embodiment of the application provides a motion trail prediction method and a computer program product of a cross-medium aircraft, which can solve the problem of insufficient precision of the motion trail of the current predicted cross-medium aircraft and improve the accuracy of the motion trail prediction of the cross-medium aircraft. In one aspect, an embodiment of the present application provides a method for predicting a motion trail of a cross-medium aircraft, including: Acquiring observation state parameters of the cross-medium aircraft at the current time point; Inputting the observation state parameters of the current time point into a trained track prediction model, outputting the prediction state parameters of the cross-medium aircraft at the next time point by adopting the trained track prediction model, wherein the trained track prediction model comprises a trained state prediction model and a trained cyclic neural network, the trained state prediction model comprises the prediction state parameters of the cross-medium aircraft at the previous time point, the trained state prediction model is used for fusing the prediction state parameters of the previous time point and the observation state parameters of the current time point to obtain corrected prediction state parameters of the current time point, calculating the prediction state of the next time point according to the corrected prediction state parameters of the current time point, and the trained cyclic neural network is used for calculating a weight coefficient between the prediction state parameters of the previous time point and the observation state parameters of the current time point during fusion; And determining the motion trail of the medium-crossing aircraft according to the predicted state parameters of the medium-crossing aircraft at the next time point. In another aspect, an embodiment of the present application provides a motion trajectory prediction apparatus for a cross-medium aircraft, including: The acquisition module is used for acquiring the observation state parameters of the cross-medium aircraft at the current time point; The prediction module is used for inputting the observation state parameters of the current time point into a trained track prediction model, outputting the prediction state parameters of the cross-medium aircraft at the next time point by adopting the trained track prediction model, wherein the trained track prediction model comprises a trained state prediction model and a trained cyclic neural network, the trained state prediction model comprises the prediction state parameters of the cross-medium aircraft at the previous time point, the trained state prediction model is used for fusing the prediction state parameters of the previous time point and the observation state parameters of the current time point to obtain corrected prediction state parameters of the current time point, calculating the prediction state of the next time point according to the corrected prediction state parameters of