Flight attitude compensation method and system for dynamic environment prediction

Abstract

The invention belongs to the field of unmanned aerial vehicles, and provides a flight attitude compensation method and system for dynamic environment prediction. The flight attitude compensation method for dynamic environment prediction comprises the following steps: processing a standard data set by using a graph neural network model to obtain a spatial-temporal characteristic graph of an environment disturbance field; modeling the environmental disturbance field through topology invariance, representing the spatial-temporal characteristic pattern of the environmental disturbance field as topology invariant characteristics, and obtaining predicted distribution of the environmental disturbance field; calculating the required attitude compensation amount according to the predicted distribution of the environmental disturbance field and the attitude demand of the aircraft, and adjusting the attitude control input of the aircraft according to the attitude compensation amount to obtain a corresponding control signal so as to realize attitude compensation; and in the flight process, the predicted distribution of the environment disturbance field is updated in real time, and the attitude compensation amount of the aircraft is dynamically adjusted through a self-adaptive control strategy according to the predicted distribution so as to stabilize the flight attitude of the aircraft. The method can effectively guarantee the precision and robustness of aircraft attitude compensation in a complex dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of unmanned aerial vehicles, and particularly relates to a flight attitude compensation method and system for dynamic environment prediction. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] During the flight of an aircraft, the aircraft is disturbed by environmental factors such as air flow, temperature change, turbulence intensity, etc., which can significantly affect the attitude stability of the aircraft. In order to ensure the safety and accuracy of the aircraft, attitude compensation technology has emerged. Traditional flight attitude compensation methods usually rely on physical model-based control strategies and simplified environmental disturbance models, but these methods have significant limitations in dealing with complex disturbances and adapting to dynamic environments.

[0004] When existing graph neural network models are applied to flight attitude compensation, although the graph neural network can learn spatio-temporal features and propagate features, the existing technology has not been able to effectively maintain the topology of the graph, i.e., topological invariance. The spatio-temporal characteristics of environmental data are key in aircraft attitude compensation, while graph neural networks usually update features in multiple convolution operations, which can cause the topological relationship between nodes to change during multiple convolution processes, thereby affecting the prediction accuracy of the disturbance field. The lack of an effective topology maintenance mechanism makes it impossible to effectively guarantee the accuracy and robustness of aircraft attitude compensation in complex dynamic environments. SUMMARY

[0005] To solve the above technical problems, the present application provides a flight attitude compensation method and system for dynamic environment prediction, which can effectively guarantee the accuracy and robustness of aircraft attitude compensation in complex dynamic environments.

[0006] To achieve the above purpose, the present application adopts the following technical solutions: The first aspect of the present application provides a flight attitude compensation method for dynamic environment prediction.

[0007] In one or more embodiments, a flight attitude compensation method for dynamic environment prediction is provided, comprising: Obtaining real-time environmental data of an aircraft and preprocessing to generate a standard data set; Using a graph neural network model to process the standard data set to obtain a spatio-temporal feature map of the environmental disturbance field; Modeling the environmental disturbance field by topological invariance, representing the spatio-temporal feature map of the environmental disturbance field as topologically invariant features to obtain a prediction distribution of the environmental disturbance field; According to the predicted distribution of the environmental disturbance field and the attitude requirement of the aircraft, a required attitude compensation amount is calculated, and an attitude control input of the aircraft is adjusted according to the attitude compensation amount, so that a corresponding control signal is obtained to realize attitude compensation. During flight, the predicted distribution of the environmental disturbance field is updated in real time, and the attitude compensation amount of the aircraft is dynamically adjusted through an adaptive control strategy to stabilize the flight attitude of the aircraft.

[0008] As an implementation mode, the process of representing the spatio-temporal feature map of the environmental disturbance field as a topologically invariant feature is as follows: The spatio-temporal feature map of the environmental disturbance field output by each graph neural network model is uniformly dimensionally transformed by using a topologically invariant disturbance field modeling network, the comparability of different types of environmental data in the feature space is constructed, and the features of the corresponding environmental data are obtained; The features of the environmental data after uniform dimensional transformation are spatially aligned to generate a spatial correspondence relationship between the features of different environmental data at the same time step; The features of each environmental data are integrated in multiple levels step by step through topological invariance; The features of different modal environmental data are sequentially fused to obtain a topologically invariant representation of the disturbance field at each time.

[0009] As an implementation mode, the features of different modal environmental data are sequentially fused by a fusion function, and the fusion process of the fusion function is as follows: The neighbor nodes of each node are aggregated by weighted summation, and the weight is determined by the adjacency matrix The adjacency matrix represents the connection relationship between nodes in the graph, and the node At time The aggregated feature is: ; wherein, The set of neighbor nodes of node is denoted as , the element in the adjacency matrix represents the connection relationship between node and node , and represents the feature vector of node j at time step t after alignment processing. The self-feature of each node is fused with the aggregated feature of its neighbor nodes, and the fusion process is completed by weighted summation to generate the fused node feature : ; wherein, is a fusion weight matrix, is a bias term, is an activation function, represents a connection operation; The fusion keeps the topological structure of the graph and the spatio-temporal structure of the environmental disturbance field.

[0010] As an embodiment, the spatio-temporal feature graph includes nodes representing data features of the aircraft's surrounding environment sensors, the feature vector of the node including values of the environmental data at the current time ; the edges in the graph represent the connection relationship between the nodes, and the connection relationship is constructed according to the spatial layout of the aircraft, the relative position of the sensors and the time sequence, and the weight of the edge is weighted based on the spatial distance.

[0011] As an embodiment, the spatio-temporal feature graph extraction process of the environmental disturbance field is: For the environmental data of each time in the standard data set, an independent graph neural network model is configured respectively; The standardized wind speed data is input into the first graph neural network model, the angular rate data is input into the second graph neural network model, the temperature data is input into the third graph neural network model, and the turbulence intensity data is input into the fourth graph neural network model; The node features in each graph neural network model are constructed into a spatio-temporal feature graph; The topological structure preserving regularization term is applied to the feature representation output by each graph neural network model respectively, so as to keep the spatio-temporal structure of the node features in the disturbance field unchanged; The feature vectors output by each graph neural network model are recorded and output respectively, and the spatio-temporal feature graph of the environmental disturbance field is obtained.

[0012] As an embodiment, the structure preserving invariance of the spatio-temporal feature graph keeps the spatial and temporal structure relationship between the nodes in the graph convolution process; the graph neural network updates and propagates the node features through the graph convolution operation, and the features of each node can be influenced by the features of its neighbor nodes, and learns the spatio-temporal evolution characteristics of the environmental disturbance field; the node features of the graph are processed through the multi-layer graph convolution network, and the spatio-temporal feature graph of the environmental disturbance field is generated.

[0013] As an embodiment, during the flight, the adaptive control strategy is used to adjust the attitude compensation amount of the aircraft according to the updated prediction distribution of the environmental disturbance field, and through the real-time calculation of the compensation amount, the error between the aircraft attitude and the target attitude is minimized.

[0014] The second aspect of the application provides a flight attitude compensation system for dynamic environment prediction.

[0015] In one or more embodiments, a flight attitude compensation system for dynamic environment prediction includes: a standard data set generation module configured to obtain real-time environment data of the aircraft and pre-process the real-time environment data to generate a standard data set; a spatio-temporal feature map construction module configured to process the standard data set using a graph neural network model to obtain a spatio-temporal feature map of the environmental disturbance field; a disturbance field prediction distribution module configured to model the environmental disturbance field by topological invariance, represent the spatio-temporal feature map of the environmental disturbance field as topologically invariant features, and obtain a prediction distribution of the environmental disturbance field; an attitude control input adjustment module configured to calculate a required attitude compensation amount according to the prediction distribution of the environmental disturbance field and an attitude requirement of the aircraft, and adjust attitude control input of the aircraft according to the attitude compensation amount to obtain a corresponding control signal to achieve attitude compensation; an attitude compensation amount dynamic adjustment module configured to update the prediction distribution of the environmental disturbance field in real time during flight, and dynamically adjust the attitude compensation amount of the aircraft through an adaptive control strategy to stabilize the flight attitude of the aircraft.

[0016] A third aspect of the present application provides a computer-readable storage medium.

[0017] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps in the dynamic environment prediction-based flight attitude compensation method described above.

[0018] A fourth aspect of the present application provides an electronic device.

[0019] An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the dynamic environment prediction-based flight attitude compensation method described above when executing the program.

[0020] Compared with the prior art, the present application has the following advantages: The application innovatively proposes a flight attitude compensation technology of dynamic environment prediction, which standardizes the aircraft environment data, extracts the space-time feature map by using a graph neural network model, effectively captures the space-time change pattern of environmental disturbances such as wind speed, temperature and turbulence intensity, in the feature modeling stage, by constructing a topological invariance maintaining mechanism, ensures that the space-time structure of the disturbance field is preserved in the processing process, avoids information loss, in the feature fusion process, the environmental data around the aircraft is learned by using graph convolution operation, and a high-precision disturbance field prediction result is generated, in the case of missing data, combined with real-time sensor data and graph neural network output, through an adaptive strategy, the missing modal data is estimated and compensated, through the adaptive control strategy, based on the real-time updated disturbance field prediction, the aircraft attitude compensation amount is dynamically adjusted, the adaptability and stability of the control system in the complex dynamic environment are improved, the problems of attitude control response lag, insufficient disturbance field modeling precision, poor data missing compensation ability and weak control system adaptability in the existing flight attitude compensation technology are solved, the precision and robustness of the aircraft attitude compensation are significantly improved, and solid technical support is provided for actual flight tasks. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this application. The embodiments of the application illustrate the

[0022] Figure 1 is a flow diagram of a dynamic environment prediction flight attitude compensation method of an embodiment of the application; Figure 2 is a structural diagram of a dynamic environment prediction flight attitude compensation system of an embodiment of the application; Figure 3 is a schematic diagram of an electronic device according to an embodiment of the application. DETAILED DESCRIPTION

[0023] The application will be further described below in conjunction with the drawings and embodiments.

[0024] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0025] It is to be understood that the terms used herein are for the purpose of describing specific embodiments and are not intended to limit exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components, and / or combinations thereof.

[0026] Most of the flight attitude compensation techniques currently use traditional control methods such as PID control, adaptive control, and robust control. These methods calculate the compensation amount by the difference between the attitude error of the aircraft and the preset target attitude, and adjust the control input of the aircraft. Traditional methods usually rely on fixed control parameters and pre-set disturbance models. In the face of complex environmental disturbances, the adaptability of traditional methods is poor, and it is difficult to meet the precise compensation requirements under rapidly changing environmental conditions.

[0027] Existing flight attitude compensation methods usually rely on a single source of environmental data or limited sensor inputs. In a complex environment, the data obtained by each sensor of the aircraft has spatial and temporal dependence, and the measurement error and response characteristics of each sensor may be different. Existing technologies rely on single-mode data for compensation calculation, ignoring the synergy between different sensors and failing to fully utilize the complementarity of multi-sensor data. This single-mode data processing method, especially in dynamic environments, shows low adaptability and instability, and cannot meet the demand for high-precision compensation.

[0028] In recent years, new technologies such as deep learning and machine learning have been gradually introduced into the field of flight attitude compensation. In particular, by using graph neural networks to process multi-modal data with spatial and temporal dependence, it is possible to more effectively capture the complex spatio-temporal structure of the environmental disturbances around the aircraft. Graph neural networks can learn the relationship between nodes through graph convolution operations and capture the spatio-temporal variation patterns of the environmental disturbance field. Therefore, using graph neural networks for spatio-temporal feature modeling has become a new solution.

[0029] Although the graph neural network can process spatio-temporal features, there are still great challenges in establishing the graph structure and fusing multi-modal data when processing different modal data. At the spatio-temporal feature fusion stage, many current methods often use simple splicing or shallow mapping, ignoring the physical differences between different modal data and the characteristics of the spatial structure, resulting in insufficient expression of the fused features and failing to fully mine the deep correlations between modalities. In actual application, the sensor data of the aircraft often has missing data, especially in multi-modal data, the output of some sensors may be missing or incomplete, and the existing graph neural network model has poor adaptability to such data missing, and cannot effectively process incomplete input, resulting in insufficient prediction stability and generalization ability of the model when facing incomplete data.

[0030] Figure 1 is a flowchart of a dynamic environment prediction flight attitude compensation method in an embodiment of the present application, as Figure 1 indicated, the dynamic environment prediction flight attitude compensation method in the embodiment can include the following steps S101-S105.

[0031] The specific implementation process of steps S101-S105 is as follows: Step S101: Obtain real-time environment data of the aircraft and perform preprocessing to generate a standard data set.

[0032] In the specific implementation process of step S101, the real-time environment data of the aircraft is obtained, including the attitude data, angular rate data, wind speed data, turbulence intensity data, and temperature data of the aircraft. The environment data is preprocessed, including but not limited to: denoising the original environment data to obtain accurate environment data; standardizing the environment data to convert the data to a value within a standard range to obtain standardized data; filling in the blank points caused by data missing through an interpolation algorithm (other existing data filling methods can also be used) to obtain a continuous environment data time series; generating a standard data set according to the processed data.

[0033] Step S102: Process the standard data set using a graph neural network model to obtain a spatio-temporal feature map of the environment disturbance field.

[0034] Specifically, the spatio-temporal feature map extraction process of the environment disturbance field is as follows: Step S1021: For each time's environment data in the standard data set, an independent graph neural network model is configured respectively; Step S1022: input the standardized wind speed data into the first graph neural network model, the angular rate data into the second graph neural network model, the temperature data into the third graph neural network model, and the turbulence intensity data into the fourth graph neural network model; Step S1023: construct the node features in each graph neural network model into a spatiotemporal feature map; Step S1024: apply a topological structure preserving regularization term to the feature representation output by each graph neural network model respectively, to keep the spatiotemporal structure of the node features in the perturbation field unchanged; Step S1025: record and output the feature vectors output by each graph neural network model respectively, to obtain the spatiotemporal feature map of the environment perturbation field.

[0035] It should be noted here that the structures of the first, second, third and fourth graph neural network models are all existing structures, which can be trained according to the corresponding input data and output spatiotemporal feature map.

[0036] In the present embodiment, the spatiotemporal feature map includes node representing the data features of the sensors around the aircraft, the feature vector of the node including the values of the environmental data at the current time ; the edges in the graph represent the connection relationship between the nodes, which is constructed according to the spatial layout of the aircraft, the relative position of the sensors and the time sequence, and the weight of the edge is weighted based on the spatial distance.

[0037] The structure preserving invariance of the spatiotemporal feature map preserves the spatial and temporal structure relationship between the nodes in the graph convolution process; the graph neural network updates and propagates the node features through the graph convolution operation, and the features of each node can be influenced by the features of its neighbor nodes, and learns the spatiotemporal evolution characteristics of the environment perturbation field; the node features of the graph are processed through the multi-layer graph convolution network, to generate the spatiotemporal feature map of the environment perturbation field.

[0038] Step S103: model the environment perturbation field through topological invariance, represent the spatiotemporal feature map of the environment perturbation field as topologically invariant features, to obtain the predicted distribution of the environment perturbation field.

[0039] Specifically, the process of representing the spatiotemporal feature map of the environment perturbation field as topologically invariant features is as follows: Step S1031: use the pre-constructed topologically invariant perturbation field modeling network to perform uniform dimension transformation on the spatiotemporal feature map of the environment perturbation field output by each graph neural network model, to build the comparability of different types of environmental data in the feature space, to obtain the features of the corresponding environmental data; Step S1032: perform spatial alignment processing on the features of the environmental data after uniform dimension transformation, to generate the spatial correspondence relationship between the features of different environmental data at the same time step; Step S1033: integrate the features of each environmental data through topological invariance in multiple levels step by step; Step S1034: sequentially fuse the features of the environment data of different modalities to obtain the topological invariant representation of the disturbance field at each time.

[0040] In step S1034, the features of the environment data of different modalities are sequentially fused by a fusion function, wherein the fusion process of the fusion function is: The neighbor nodes of each node are aggregated by weighted summation, and the weight is determined by the adjacency matrix provided, the adjacency matrix represents the connection relationship between nodes in the graph, and the node At time The aggregated feature of the node is: ; wherein, represents the neighbor node set of the node , and is an element in the adjacency matrix, representing the connection relationship between the node and the node , and represents the feature vector of the node j at time step t after alignment processing. The feature of each node itself is fused with the aggregated feature of its neighbor nodes , and the fusion process is completed by weighted summation to generate the fused node feature : ; wherein, is a fusion weight matrix, is a bias term, is an activation function, denotes a connection operation. The fusion preserves the topological structure of the graph and the spatio-temporal structure of the environmental disturbance field.

[0041] The node feature obtained by the multi-layer graph convolution operation contains the environmental data features at the current time and preserves the topological relationship between the nodes in the graph.

[0042] Step S104: according to the predicted distribution of the environmental disturbance field and the attitude requirement of the aircraft, the required attitude compensation amount is calculated, and the attitude control input of the aircraft is adjusted accordingly to obtain the corresponding control signal to realize attitude compensation.

[0043] Specifically, according to the predicted distribution of the environmental disturbance field, the predicted disturbance field features are combined with the attitude requirements of the aircraft to calculate the required attitude compensation amount; by predicting the environmental disturbance field, the predicted values of wind speed, angular rate, temperature and turbulence intensity are obtained, and these predicted values are compared with the target attitude requirements of the aircraft; according to the calculated attitude compensation amount, the attitude control input of the aircraft is adjusted through the control system of the aircraft, and the corresponding control signal is output to realize attitude compensation.

[0044] Step S105: During flight, the predicted distribution of the environmental disturbance field is updated in real time, and the attitude compensation amount of the aircraft is dynamically adjusted through the adaptive control strategy to stabilize the flight attitude of the aircraft.

[0045] Wherein, during flight, the attitude compensation amount of the aircraft is adjusted using the adaptive control strategy according to the updated predicted distribution of the environmental disturbance field, and through real-time calculation of the compensation amount, the error between the aircraft attitude and the target attitude is minimized.

[0046] In the aircraft control system, real-time feedback is combined to update the control input to dynamically adapt to environmental changes and ensure the attitude stability of the aircraft in complex and dynamic environments; during control, gain scheduling adaptive strategy is implemented according to the changes in the disturbance field and the actual performance of the aircraft to adjust the attitude of the aircraft and stabilize the flight attitude of the aircraft.

[0047] In a UAV flight task, the UAV is deployed to perform a mountain reconnaissance task. The task requires the UAV to fly stably in a complex airflow change and turbulence environment, ensuring that the flight attitude does not fluctuate greatly. The task flight area has complex terrain, and is accompanied by strong winds all year round. The wind speed changes sharply during flight, the turbulence intensity is high, and the airflow is unstable, which brings challenges to the flight attitude control of the UAV. Traditional flight attitude compensation methods, such as PID control or adaptive control, often cannot respond to rapid changes in the environment in time, resulting in large fluctuations in the flight attitude and failing to meet the demand for precise and stable flight.

[0048] Traditional flight attitude compensation methods usually rely on linear models and fixed control strategies, such as PID control and adaptive control. These methods usually use simple sensor data input and assume that the environmental disturbances experienced by the aircraft are known or can be predicted through a simple model. In actual applications, the environment in which the aircraft is located is complex and dynamic, and the changes in airflow, temperature, turbulence and other environmental factors are nonlinear and unpredictable. Traditional methods lack real-time perception and prediction capabilities for these complex and rapidly changing environmental disturbances, resulting in insufficient control accuracy of the aircraft attitude, especially when facing complex environments such as strong winds and turbulence, the adaptability and robustness of traditional control systems decrease significantly.

[0049] The method of the present application introduces dynamic prediction capability for the aircraft environment through a graph neural network model. The graph neural network can model the spatio-temporal features of environmental disturbances through graph convolution operations, capturing the spatial and temporal dependencies of these environmental factors. In a dynamic environment, it can accurately predict the changes in environmental disturbances in real time and adjust the aircraft attitude compensation based on these predictions, ensuring the stability of the aircraft attitude.

[0050] In flight missions, unmanned aerial vehicles are equipped with various sensors such as anemometers, thermometers, gyroscopes, accelerometers, etc. to collect real-time data in multiple dimensions of the environment in which the aircraft is located. These data include wind speed, angular rate, temperature, and turbulence intensity, and the sensors collect data every second. Through the multi-layer convolution operation of the graph neural network, the environmental data is mapped into a graph structure, with nodes representing the environmental data points around the aircraft and edges representing their spatial or temporal dependencies. This structure enables the aircraft to obtain comprehensive spatio-temporal feature information in a complex environment, which is propagated through the convolution operation of the graph neural network, enabling the aircraft to accurately predict environmental changes in the future. Unlike traditional control methods that rely solely on simple mathematical models, the graph neural network can capture more complex spatio-temporal relationships, resulting in more accurate predictions, especially when facing unknown or sudden environmental disturbances, providing faster responses and more accurate compensation.

[0051] In traditional graph neural network models, information propagation between nodes during graph convolution can cause changes in the topological relationship between nodes, affecting feature expression and prediction results. In the present application, a topology structure preservation regularization term is introduced to ensure that the spatial structure between nodes is preserved during graph convolution. This mechanism addresses the problem of loss of environmental data structure in traditional methods, enabling the graph neural network to maintain the spatial relationship of the environment around the aircraft after multi-layer convolution, thereby improving the accuracy of spatio-temporal feature modeling and compensation effect.

[0052] In the experiment, the target attitude of the aircraft was set to stable flight, with a target angle of zero, and the aircraft needed to maintain a certain trajectory for reconnaissance tasks. When the wind speed suddenly changed from 4 m / s to 10 m / s, the maximum attitude error of the aircraft using the traditional PID control method was 6°, with significant attitude fluctuations. However, using the graph neural network method based on the present application, the attitude error of the aircraft was always maintained within 1°, and the aircraft could adjust its attitude in time and maintain near the target angle in the strong wind area. The experiment also showed that when the aircraft encountered changes in turbulence intensity, the compensation effect of the traditional PID control method decreased significantly, while the method of the present application could quickly adjust the compensation amount to ensure the stability of the aircraft attitude with high compensation accuracy.

[0053] Table 1 Comparison of the method of the present application and the conventional method

[0054] The present application applies a graph neural network to the field of flight attitude compensation and combines a topological invariance preservation mechanism to solve the shortcomings of existing flight attitude control methods in dynamic environments, significantly improving the attitude control stability and precision of the aircraft, and overcoming the problems of slow response to complex environmental disturbances and insufficient prediction accuracy of traditional methods. By predicting environmental disturbances in real time and dynamically adjusting the compensation amount, the present application provides an accurate, flexible and efficient attitude compensation method for the aircraft, ensuring safe flight of the aircraft in complex environments.

[0055] As Figure 2 shown, the flight attitude compensation system for dynamic environment prediction provided by the embodiments of the present application can be implemented in a software manner, and the flight attitude compensation system for dynamic environment prediction includes the following software modules: a standard data set generation module 201, a spatiotemporal feature map construction module 202, a disturbance field prediction distribution module 203, an attitude control input adjustment module 204, and an attitude compensation amount dynamic adjustment module 205.

[0056] The functions of each software module in the flight attitude compensation system for dynamic environment prediction will be introduced as follows: The standard data set generation module 201 is used to obtain real-time environmental data of the aircraft and perform preprocessing to generate a standard data set; The spatiotemporal feature map construction module 202 is used to process the standard data set using a graph neural network model to obtain a spatiotemporal feature map of the environmental disturbance field; The disturbance field prediction distribution module 203 is used to model the environmental disturbance field through topological invariance, express the spatiotemporal feature map of the environmental disturbance field as topologically invariant features, and obtain a prediction distribution of the environmental disturbance field; The attitude control input adjustment module 204 is used to calculate the required attitude compensation amount according to the prediction distribution of the environmental disturbance field and the attitude requirements of the aircraft, and adjust the attitude control input of the aircraft accordingly to obtain a corresponding control signal to achieve attitude compensation; The attitude compensation amount dynamic adjustment module 205 is used to update the prediction distribution of the environmental disturbance field in real time during flight, and dynamically adjust the attitude compensation amount of the aircraft through an adaptive control strategy to stabilize the flight attitude of the aircraft.

[0057] It should be noted that each module in the embodiments of the present application corresponds to each step in the flight attitude compensation method for dynamic environment prediction described above, and the specific implementation process is the same, which will not be described here again.

[0058] The application introduces a graph neural network (GNN) and a topological invariance maintenance mechanism, and proposes a flight attitude compensation method based on dynamic environment prediction to solve the problems of slow response of aircraft attitude control in a complex dynamic environment, insufficient modeling of temporal and spatial changes of disturbance fields, poor processing of missing data, and poor adaptability of the control system. In the feature modeling stage, the topological invariance maintenance mechanism is constructed to ensure that the temporal and spatial structure of the disturbance field is preserved during processing, avoiding information loss.

[0059] In the feature fusion process, the environmental data around the aircraft is learned using graph convolution operations to generate high-precision disturbance field prediction results. Figure 3 In the case of missing data, real-time sensor data and graph neural network output are combined to estimate and compensate for missing modal data through an adaptive strategy. Figure 3 Through an adaptive control strategy, the disturbance field prediction is updated in real time, and the aircraft attitude compensation amount is dynamically adjusted to improve the adaptability and stability of the control system in a complex dynamic environment.Overall, the method effectively overcomes the problems of attitude control response lag, insufficient disturbance field modeling accuracy, poor data missing compensation ability, and weak adaptability of the control system in existing flight attitude compensation technology, significantly improving the precision and robustness of aircraft attitude compensation, and providing solid technical support for actual flight missions.

[0060] The electronic device provided by the embodiment of the application includes at least one processor 301, a memory 302, a user interface 303, and at least one network interface 304. Figure 3 The various buses in the bus system 305 are used to realize the connection and communication between the components.

[0061] The user interface 303 can include a display, a keyboard, a mouse, a trackball, a click wheel, a key, a button, a touchpad, or a touch screen, etc.

[0062] ​​It is to be understood that the memory 302 can be volatile or non-volatile memory, and can also include both volatile and non-volatile memory. The memory 302 in the embodiments of the present application can store data to support the operation of the terminal. Examples of these data include any computer programs for operating on the terminal, such as an operating system and application programs. Among them, the operating system contains various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application programs can include various application programs.

[0063] In some embodiments, the dynamic environment prediction flight attitude compensation system provided by the embodiments of the present application can be implemented in a combination of software and hardware. As an example, the dynamic environment prediction flight attitude compensation system provided by the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the dynamic environment prediction flight attitude compensation method provided by the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can use one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components.

[0064] As an example, the processor 301 can be an integrated circuit chip with signal processing capability, such as a general purpose processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc., wherein the general purpose processor can be a microprocessor or any conventional processor.

[0065] As an example of the dynamic environment prediction flight attitude compensation system provided by the embodiments of the present application is implemented by hardware, the device provided by the embodiments of the present application can be directly implemented by a processor 301 in the form of a hardware decoding processor to complete the execution, for example, one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD), field programmable gate array (FPGA) or other electronic components to implement the dynamic environment prediction flight attitude compensation method provided by the embodiments of the present application.

[0066] The memory 302 in the embodiments of the present application is used to store various types of data to support the operation of the dynamic environment prediction flight attitude compensation system, or store program codes for executing the method shown in the embodiments of the present application. Figure 1 Examples of these data include any executable instructions for operating on the dynamic environment prediction flight attitude compensation system, such as executable instructions, programs for implementing the dynamic environment prediction flight attitude compensation method of the embodiments of the present application can be included in the executable instructions.

[0067] In particular, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program codes for executing the method shown in the embodiments of the present application. Figure 1 In such embodiments, the computer program can be downloaded and installed from the network by the communication part, and / or installed from the detachable medium. When the computer program is executed by the central processing unit, various functions defined in the device of the present application are executed.

[0068] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general purpose computer, a special purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a machine that implements the flow Figure 1 flow or multiple flows and / or blocks Figure 1 Figure 1means for performing the function specified in the block or blocks.

[0069] The above descriptions are merely some embodiments of the present application, but are not intended to limit the present application. The present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall fall within the scope of the present application.

Claims

1. A method of flight attitude compensation for dynamic environment prediction, characterized in that, The method comprises the following steps: acquiring real-time environment data of the aircraft and preprocessing the data to generate a standard data set; processing the standard data set using a graph neural network model to obtain a spatio-temporal feature map of the environmental disturbance field; modeling the environmental disturbance field through topological invariance, representing the spatio-temporal feature map of the environmental disturbance field as topologically invariant features to obtain a prediction distribution of the environmental disturbance field; calculating the required attitude compensation amount according to the prediction distribution of the environmental disturbance field and the attitude requirement of the aircraft, and adjusting the attitude control input of the aircraft accordingly to obtain a corresponding control signal to achieve attitude compensation; in the flight process, the prediction distribution of the environmental disturbance field is updated in real time, and the attitude compensation amount of the aircraft is dynamically adjusted through an adaptive control strategy to stabilize the flight attitude of the aircraft.

2. A method of flight attitude compensation for dynamic environmental prediction as claimed in claim 1, characterized in that, The process of representing the spatio-temporal feature map of the environmental disturbance field as topologically invariant features comprises the following steps: using a pre-constructed topologically invariant disturbance field modeling network to perform uniform dimension transformation on the spatio-temporal feature map of the environmental disturbance field output by each graph neural network model, building comparability of different types of environmental data in the feature space, and obtaining the features of the corresponding environmental data; performing spatial alignment processing on the features of the environmental data after uniform dimension transformation to generate a spatial correspondence relationship between the features of different environmental data at the same time step; integrating the features of each environmental data through topological invariance in multiple levels step by step; fusing the features of environmental data of different modalities in sequence to obtain a topologically invariant representation of the disturbance field at each time.

3. A dynamic environmental prediction flight attitude compensation method according to claim 2, wherein, The features of environmental data of different modalities are fused in sequence through a fusion function, and the fusion process of the fusion function is as follows: For each node, the neighboring nodes are summed and their information is aggregated using a weighted summation method. The weights are determined by the adjacency matrix. Provides an adjacency matrix. This represents the connection relationship between nodes in the graph. At any moment aggregation features for: ;in, Represents a node The set of neighboring nodes, The elements in the adjacency matrix represent nodes. and nodes The connection between them This represents the feature vector of node j at time step t after alignment. aggregating features of each node with its neighbor nodes performing a fusion, the fusion being done by weighted sum, generating fused node features :​ ； wherein, is a fusion weight matrix, is a bias term, is an activation function, denotes a concatenation operation; the fusion preserves the topological structure of the graph and the spatio-temporal structure of the environmental disturbance field.

4. The method of dynamic environmental prediction of flight attitude compensation as claimed in claim 1, wherein, Spatiotemporal feature graphs include nodes The feature vector of a node represents the data characteristics of sensors in the environment surrounding the aircraft, and includes environmental data at the current moment. The value of ; the edges in the graph represent the connection relationship between nodes, which is constructed based on the spatial layout of the aircraft, the relative position of the sensors and the time sequence, and the weight of the edges is weighted based on spatial distance.

5. The method of dynamic environmental prediction of flight attitude compensation as claimed in claim 1, wherein, The structure of the spatio-temporal feature map remains unchanged, and the spatial and temporal structure relationship between nodes is preserved in the graph convolution process; the graph neural network updates and propagates the node features through graph convolution operation, and each node's feature can be influenced by its neighbor node's feature, and learns the spatio-temporal evolution characteristics of the environmental disturbance field; the node features of the graph are processed through a multi-layer graph convolution network to generate a spatio-temporal feature map of the environmental disturbance field.

6. A method of flight attitude compensation for dynamic environmental prediction as claimed in claim 1, wherein, The spatio-temporal feature map extraction process of the environmental disturbance field comprises the following steps: for each time's environmental data in the standard data set, an independent graph neural network model is configured respectively; the standardized wind speed data is input into the first graph neural network model, the angular rate data is input into the second graph neural network model, the temperature data is input into the third graph neural network model, and the turbulence intensity data is input into the fourth graph neural network model; the node features in each graph neural network model are constructed into a spatio-temporal feature map; topological structure preserving regularization terms are applied to the feature representation output by each graph neural network model to maintain the spatio-temporal structure of the node features in the disturbance field unchanged; the feature vectors output by each graph neural network model are recorded and output respectively to obtain a spatio-temporal feature map of the environmental disturbance field.

7. A method of flight attitude compensation for dynamic environmental prediction as claimed in claim 1, wherein, In the flight process, the attitude compensation amount of the aircraft is adjusted using an adaptive control strategy according to the updated prediction distribution of the environmental disturbance field, and the real-time calculation of the compensation amount ensures that the error between the aircraft attitude and the target attitude is minimized.

8. A dynamic environment predicted flight attitude compensation system, characterized in that, The method comprises the following steps: a standard dataset generation module configured to obtain real-time environment data of the aircraft and pre-process the real-time environment data to generate a standard dataset; a spatio-temporal feature map construction module configured to process the standard dataset using a graph neural network model to obtain a spatio-temporal feature map of the environmental disturbance field; a disturbance field prediction distribution module configured to model the environmental disturbance field by topological invariance, represent the spatio-temporal feature map of the environmental disturbance field as topologically invariant features, and obtain a prediction distribution of the environmental disturbance field; an attitude control input adjustment module configured to calculate a required attitude compensation amount according to the prediction distribution of the environmental disturbance field and an attitude requirement of the aircraft, and adjust attitude control input of the aircraft according to the required attitude compensation amount to obtain a corresponding control signal to achieve attitude compensation; an attitude compensation amount dynamic adjustment module configured to update the prediction distribution of the environmental disturbance field in real time during flight, and dynamically adjust the attitude compensation amount of the aircraft through an adaptive control strategy to stabilize the flight attitude of the aircraft.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps in the dynamic environment prediction flight attitude compensation method of any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps in the dynamic environment prediction flight attitude compensation method of any one of claims 1-7.

Images