CN-121278577-B - Unmanned aerial vehicle flight state simulation method based on digital twin

CN121278577BCN 121278577 BCN121278577 BCN 121278577BCN-121278577-B

Abstract

The invention relates to the technical field of unmanned aerial vehicle simulation, and discloses a digital twinning-based unmanned aerial vehicle flight state simulation method. The method comprises the steps of collecting multi-source flight data streams such as position information, motion parameters, environment perception data and the like of a physical unmanned aerial vehicle by constructing a digital twin system comprising a physical unmanned aerial vehicle entity and a virtual simulation model. And carrying out multidimensional feature fusion processing on the multi-source data to generate fusion feature representation, and determining a flight state estimated value and related confidence information by adopting a probabilistic state modeling algorithm based on the fusion feature. And updating the virtual simulation model according to the estimated value and the confidence coefficient to simulate the real-time flight state. A flight status cluster analysis is performed in the virtual model to detect abnormal patterns and an adaptive simulation threshold set is generated based on historical flight data. And comparing the abnormal mode with the self-adaptive threshold set, and outputting a simulation control instruction for unmanned aerial vehicle operation. The method can effectively support accurate simulation and abnormal control of the unmanned aerial vehicle flight state.

Inventors

ZHANG ZHENYI
YAN CHUANBO
HU TAO
ZHAO YIN

Assignees

辽宁省航科信创科技发展有限公司

Dates

Publication Date: 20260505
Application Date: 20250829

Claims (9)

1. The unmanned aerial vehicle flight state simulation method based on digital twinning is characterized by comprising the following steps of: Constructing an unmanned aerial vehicle digital twin system, wherein the unmanned aerial vehicle digital twin system comprises a physical unmanned aerial vehicle entity and a corresponding virtual simulation model; collecting a multi-source flight data stream of the physical unmanned aerial vehicle entity, wherein the multi-source flight data stream comprises position information, motion parameters and environment perception data; Performing multidimensional feature fusion processing on the multi-source flight data stream to generate fusion feature representation; Based on the fused feature representation, determining a flight state estimation value and related confidence information through a probabilistic state modeling algorithm, including: Inputting the fusion feature representation into a depth feature transformation network, and obtaining a depth feature expression through nonlinear mapping; constructing a probability regression model, and capturing a nonlinear relation of the flight state by using a composite kernel function; calculating a predicted distribution result of the depth feature expression, wherein the predicted distribution result comprises mean output and variance output; Quantifying the uncertainty of the model through an approximate inference algorithm, and outputting the flight state estimated value and related confidence information; The implementation of the probabilistic state modeling algorithm starts from the construction of a depth feature transformation network, the depth feature transformation network adopts a five-layer full-connection structure, an input layer receives 128-dimensional fusion feature representation, three hidden layers are respectively provided with 256, 128 and 64 neurons, an output layer generates 32-dimensional depth feature representation, each layer adopts a ReLU activation function to carry out feature scaling on the last layer of lamination LayerNormalization, a historical flight data set is used in a training stage, feature reconstruction loss is minimized through an Adam optimizer, the learning rate is set to be 0.001 and cosine annealing strategy is adopted for adjustment, a network parameter update period is synchronous with the state sampling rate of the unmanned plane, forward propagation calculation is carried out every 200 milliseconds, a residual connection mechanism is introduced in the calculation process of the depth feature representation, the output of a second hidden layer and the input feature are weighted and summed, and the stability of feature transmission is enhanced; Updating the virtual simulation model to simulate a real-time flight state according to the flight state estimated value and the relevant confidence information; In the virtual simulation model, performing a flight status cluster analysis to detect an abnormal pattern; generating an adaptive simulation threshold set based on the historical flight data record; Comparing the abnormal mode with the self-adaptive simulation threshold set, and outputting a simulation control instruction for unmanned aerial vehicle operation.
2. The unmanned aerial vehicle flight state simulation method based on digital twinning according to claim 1, wherein the multi-dimensional feature fusion processing is performed on the multi-source flight data stream, and the method comprises: carrying out noise characteristic evaluation on the multi-source flight data stream, and determining a data denoising threshold parameter; Performing filtering pretreatment on the multi-source flight data stream according to the data denoising threshold parameter to obtain a standard multi-source flight data stream; Extracting a space-time correlation feature set of the standard multi-source flight data stream; setting a characteristic weight dynamic allocation scheme based on the sensor reliability index; and carrying out weighted fusion calculation on the space-time associated feature set according to the feature weight dynamic allocation scheme to generate the fusion feature representation.
3. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 1, wherein in the virtual simulation model, performing a flight state cluster analysis to detect an anomaly pattern comprises: Inputting current flight state data into a density-based clustering engine; Identifying outliers in the clustering results output by the density-based clustering engine; Calculating an abnormal metric value of the outlier, and implementing dynamic threshold adjustment based on local neighborhood distance analysis; and determining the abnormal mode according to the abnormal measurement value.
4. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 1, wherein the generating an adaptive simulation threshold set based on historical flight data records comprises: Extracting a normal operating condition data set from the historical flight data record; training the generated model to simulate the normal flight state characteristics; dynamically learning state boundary features through a discriminant model; generating the multi-level threshold set includes a low risk threshold, a medium risk threshold, and a high risk threshold.
5. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 1, wherein the method further comprises: constructing a distributed processing architecture, the distributed processing architecture comprising a plurality of computing nodes; Adaptively distributing processing tasks of the multi-source flight data stream to the plurality of computing nodes based on real-time network conditions and load conditions.
6. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 5, wherein the adaptively distributing processing tasks of the multi-source flight data stream based on real-time network conditions and load conditions comprises: Setting task allocation rules comprising a bandwidth allocation mechanism and a routing mechanism; Constructing a strategy optimization space based on the historical transmission record; and (5) matching the real-time constraint conditions and determining a node task allocation scheme.
7. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 1, wherein the output simulation control instructions comprise: performing security processing on the simulation control instruction by applying an encryption processing algorithm; and transmitting the simulation control instruction after the safety processing to the unmanned aerial vehicle execution unit.
8. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 1, wherein the method further comprises: Setting an analog interference signal source, and modulating interference signal parameters according to the flight state estimated value; injecting the disturbance signal parameters into the multi-source flight data stream to verify simulation robustness.
9. The digital twinning-based unmanned aerial vehicle flight state simulation method of claim 2, wherein the extracting the spatio-temporal correlation feature set of the standard multi-source flight data stream comprises: Dividing the standard multi-source flight data stream into a time slice sequence; calculating a statistical feature vector and a frequency domain feature vector of each time segment sequence; combining the statistical feature vector and the frequency domain feature vector to form the spatio-temporal associated feature set.

Description

Unmanned aerial vehicle flight state simulation method based on digital twin Technical Field The invention relates to the technical field of unmanned aerial vehicle simulation, in particular to a digital twinning-based unmanned aerial vehicle flight state simulation method. Background Along with the continuous expansion of unmanned aerial vehicle application scene, from civil aerial photography, commodity circulation transportation to industry inspection, military reconnaissance, the accurate control of unmanned aerial vehicle flight condition becomes the key of guarantee flight safety and task efficiency increasingly. In the development, testing and actual running processes of the unmanned aerial vehicle, the flight state simulation technology plays an important role, and the flight process of the unmanned aerial vehicle is simulated by constructing a virtual environment, so that support is provided for performance optimization, fault investigation and operation training. The traditional unmanned aerial vehicle flight state simulation method depends on a preset mathematical model and fixed parameters, and is difficult to truly reflect the complex and changeable actual flight environment. In actual flight, the flight state of an unmanned aerial vehicle is affected by a plurality of factors, including airflow disturbance, terrain variation, performance fluctuation of equipment itself and the like, and the factors interact with each other, so that the flight state presents high dynamic and uncertainty. The traditional simulation method is lack of effective integration of real-time environment data, so that a large deviation exists between a simulation result and an actual flight state, and the actual performance of the unmanned aerial vehicle in a complex environment cannot be accurately simulated. Existing simulation techniques have limitations in terms of data processing. The unmanned aerial vehicle can generate a large amount of multi-source data in the flight process, such as GPS positioning information, motion parameter data of an accelerometer, a gyroscope and the like, environment perception data of a visual sensor, a radar and the like. These data sources are different, formats are different, and noise and redundant information may be present. The traditional method mostly adopts a single dimension analysis or simple splicing mode for processing the multi-source data, and cannot realize effective feature fusion, so that effective information in the data cannot be fully mined, and further the accuracy of flight state estimation is affected. In the aspect of abnormal state detection, the prior art generally adopts a fixed threshold value for judgment, and lacks of deep utilization of historical flight data of the unmanned aerial vehicle. Because of the differences of flight characteristics of different unmanned aerial vehicles in different task scenes, the fixed threshold is difficult to adapt to diversified requirements, the condition of missed detection or false detection is easy to occur, and potential flight risks cannot be found in time. In addition, the lack of synchronism between the virtual simulation model and the physical unmanned aerial vehicle entity is also a prominent problem in the prior art, and the update of the virtual model is delayed from the state change of the physical entity, so that the simulation process cannot reflect the actual flight state of the unmanned aerial vehicle in real time, and the reliability and practicality of the simulation are reduced. Disclosure of Invention The invention aims to provide a digital twinning-based unmanned aerial vehicle flight state simulation method to solve the problems in the background technology. In order to achieve the above object, the present invention provides a digital twin-based unmanned aerial vehicle flight state simulation method, which comprises: Constructing an unmanned aerial vehicle digital twin system, wherein the unmanned aerial vehicle digital twin system comprises a physical unmanned aerial vehicle entity and a corresponding virtual simulation model; collecting a multi-source flight data stream of the physical unmanned aerial vehicle entity, wherein the multi-source flight data stream comprises position information, motion parameters and environment perception data; Performing multidimensional feature fusion processing on the multi-source flight data stream to generate fusion feature representation; Based on the fusion characteristic representation, determining a flight state estimated value and related confidence information through a probabilistic state modeling algorithm; Updating the virtual simulation model to simulate a real-time flight state according to the flight state estimated value and the relevant confidence information; In the virtual simulation model, performing a flight status cluster analysis to detect an abnormal pattern; generating an adaptive simulation threshold set based on the historical flight data