Unmanned aerial vehicle cluster digital twin model correction method based on parameter identification

Abstract

The invention discloses an unmanned aerial vehicle cluster digital twin model pose correction method based on parameter identification, belongs to the field of system modeling and simulation, and can effectively correct a digital twin model of an unmanned aerial vehicle cluster. The method comprises the following steps: firstly, establishing a digital twin model of an unmanned aerial vehicle cluster, wherein the digital twin model comprises a physical entity, a task control system, a flight control system and a physical simulation engine of an unmanned aerial vehicle; the physical entity part is used for simulating the structure and physical attributes of the unmanned aerial vehicle; the task control system is responsible for task planning and distribution; the flight control system is used for realizing attitude, speed and position control of the unmanned aerial vehicle; the physical simulation engine is used for simulating the behavior of the unmanned aerial vehicle and the influence of external environment conditions on the unmanned aerial vehicle. The method is based on a parameter identification technology, utilizes actual measurement data to dynamically update model parameters, and continuously corrects parameters and states of a twin model by identifying measurement noise and model deviation so as to realize high-precision correction of the model.

Description

Technical Field

[0001] The present invention belongs to the field of system modeling and simulation, and in particular relates to a correction method for unmanned aerial vehicle cluster digital twin modeling. Background Art

[0002] As a highly flexible flying platform, drones have the advantages of low cost, convenient operation and high mobility, and are gradually being widely used in military and civilian fields. The emergence of drone swarm technology has further magnified this advantage. Through cluster collaborative operation, drones can perform complex tasks more effectively, especially in scenarios such as environmental monitoring, disaster relief, logistics and transportation, and air patrols. Compared with traditional single drone operations, drone swarms have higher task completion efficiency and task robustness, and have significant advantages in dealing with dynamic and complex environments.

[0003] In the development and application of drone swarm systems, numerical simulation, semi-physical simulation and physical experiments are traditional verification and testing methods. However, these methods have certain limitations and are difficult to fully meet the needs of drone swarm systems in complex tasks such as intelligent decision-making and collaborative control. Numerical simulation usually uses computer models to simulate the dynamic behavior and environmental factors of drones, which has the advantages of low cost and flexible operation. However, numerical simulation is difficult to fully reflect the complexity and emergencies of the actual environment, especially in dynamic changing situations such as complex climate and multiple obstacles, the simulation accuracy and reliability are greatly limited. This limitation makes it difficult for the experimental data of numerical simulation to be fully applicable to the real world, affecting the applicability of drone swarm algorithms. The semi-physical simulation method combines computer simulation with some physical equipment, which can improve the realism of the simulation to a certain extent, such as simulating the flight control system of drones through hardware-in-the-loop technology. However, semi-physical simulation still faces the problem of high cost, and is limited by the capabilities of simulation equipment, making it difficult to dynamically reproduce the changing real environment. In addition, semi-physical simulation usually involves a relatively simplified environmental model, which cannot cover the diverse needs of large-scale drone swarms. Physical experiments directly test drones in real environments, restoring the actual operation scenarios of drones as much as possible to obtain the most realistic experimental data. However, this method has huge cost, time and safety risks, especially in bad weather or complex terrain, the cost and risk of experiments increase exponentially. In addition, physical experiments are inefficient in dynamically adjusting drone strategies and verifying new algorithms, and are not suitable for large-scale, high-frequency drone swarm intelligent decision-making training needs.

[0004] The emergence of digital twin technology provides a new approach to address the limitations of these traditional methods. Digital twins enable real-time interaction between the physical world and the digital world by building a virtual model that is highly consistent with the physical drone and its operating environment. Digital twins can not only dynamically reproduce the complex factors of the environment, but also continuously collect and update the operating data of the drone, and monitor and optimize the decision-making process and system status of the drone cluster in real time. Through the digital twin model, researchers can conduct a large number of experiments and optimizations safely and at low cost in a virtual environment, providing high-precision data support for the intelligent decision-making training of drone clusters. The rise of this technology has greatly expanded the depth and breadth of drone simulation and testing, and has become a key breakthrough in addressing the limitations of traditional simulation and experiments.

[0005] Although digital twin technology can simulate the behavior and state of drone clusters in a virtual environment with high precision, the digital twin model may produce deviations due to gradual changes in the physical system during long-term use. For example, as the components of the drone age and the environmental conditions change slightly, the gap between the model and the real system will gradually widen, thus affecting the accuracy of the simulation results. In addition, there are many influencing factors in the real environment that are difficult to observe directly, such as micro-meteorological conditions and hidden structural losses. These factors will have a subtle but important impact on the actual performance of the drone, further exacerbating the gap between the digital twin model and the physical system. Therefore, in order to ensure the long-term accuracy and effectiveness of the digital twin model, the key is to continuously calibrate it to continuously narrow the gap between virtual and real. Therefore, the key to the digital twin model lies in the continuous calibration process, which continuously introduces real data to correct and update the model to ensure its high accuracy and real-time performance in a dynamic environment. Continuous calibration based on real data can not only reduce the deviation of the model, but also improve the sustainability of the model, enabling it to adapt to the long-term evolution of the environment and system. This data-driven calibration method has become an important research direction for improving the accuracy and practicality of digital twin models, and is of great significance to the intelligent decision-making and operation efficiency of drone clusters. Summary of the invention

[0006] The purpose of the present invention is to propose a method for calibrating a drone cluster digital twin model based on parameter identification for a drone cluster digital twin simulation system. To achieve this purpose, the steps adopted by the present invention are:

[0007] Step 1: Establish a UAV digital twin model, which includes four modules: UAV cluster entity, mission control system, flight control system and physical simulation engine. The UAV cluster entity is equipped with sensors to collect flight data and transmit it back to the flight control system and mission control system. The flight control system uses ArduPilot flight control software to control both physical UAVs and virtual UAVs. The mission control system issues mission instructions and waypoint information. The physical simulation engine uses Gazebo to simulate the physical behavior of UAVs in a virtual environment.

[0008] Step 2: Establish a discrete time linear space model of the UAV digital twin model using X k-1 represents the system state at time k-1, C k-1 represents the control input at time k-1, M(·) represents the state transfer equation, B(·) represents the control equation, and the ideal state of the system at time k is expressed as:

[0009] X k =M(X k-1 )+B(C k-1 ) (1)

[0010] The actual state of the system at time k is expressed as:

[0011] X k =M(X k-1 )+B(C k-1 )+u k (2)

[0012] Among them, u k Represents the model error, and the observed value of the system is expressed as:

[0013] Z k =H(X k )+v k (3)

[0014] Where H(·) represents the measurement equation, v k represents the measurement error, Z k Represents the system observation value at time k; use represents the simulated observation value, represents the simulation state value, and the optimization model of the correction problem is expressed as:

[0015]

[0016] Among them, u opt represents the optimal estimate of the model error, v opt represents the optimal estimate of the measurement noise;

[0017] Step 3: Establish the measurement model of the UAV digital twin model. The three-axis accelerometer measurement model is expressed as:

[0018]

[0019] in B a m Represents the measurement value of the three-axis accelerometer in the body coordinate system, B a represents the true value of the linear acceleration in the body coordinate system B b a Represents the linear acceleration measurement deviation, and its first-order differential is Gaussian white noise n ba , n a Represents a Gaussian white noise matrix with a dimension of 3×1;

[0020] The three-axis gyroscope measurement model is expressed as:

[0021]

[0022] in: B ω m Represents the measurement value of the three-axis angular velocity meter in the body coordinate system, B ω represents the true value of the angular velocity in the body coordinate system, B b ω Represents the angular velocity measurement deviation, and its first-order differential is Gaussian white noise n bω , n ω Represents a Gaussian white noise matrix with a dimension of 3×1;

[0023] The barometer and GPS measurement model is expressed as:

[0024]

[0025] in N p m The measured value representing the UAV's NED position, N p represents the true value of the NED position, N b p Represents the position measurement deviation, and its first-order differential is Gaussian white noise n bp ;n p Represents a Gaussian white noise matrix with a dimension of 3×1;

[0026] Step 4: Establish the state transition model of the drone digital twin model, drone, state X at any time t It is expressed as:

[0027]

[0028] in, Nv represents the three-dimensional velocity of the drone in the north-east coordinate system, q represents the attitude of the drone expressed by quaternion at the current moment, and its leading subscript N and leading superscript B represent the rotation direction of the north-east coordinate system relative to the body coordinate system. The discrete differential state transfer equation of the drone digital twin model is expressed as:

[0029]

[0030] in, represents quaternion multiplication, q{·} represents the conversion function from angular velocity to quaternion, N g represents the gravitational acceleration vector, R represents the rotation matrix, and Δt represents the time interval;

[0031] Step 5: Parameter identification based on the least squares method is used to estimate the sensor measurement error in the initialization phase before the drone takes off. The state vector of the sensor initialization phase It is expressed as:

[0032]

[0033] A series of measurements during initialization Affected by the additive sensor error Impact:

[0034]

[0035] The mean of the additive error is calculated as:

[0036]

[0037] The variance of the additive error is calculated as:

[0038]

[0039] The model bias is estimated during the flight of the UAV. The discrete equation of the rigid body motion model of the UAV is expressed as:

[0040]

[0041] The state vector of the drone at time k is:

[0042]

[0043] The model deviation of the drone at time k is:

[0044]

[0045] where ε p，k Represented by the three-dimensional position deviation, ε v，k represents the three-dimensional velocity deviation; ε a，kRepresents the three-dimensional acceleration deviation, ε ω，k Represents the three-dimensional angular velocity deviation, and the loss function is defined as:

[0046]

[0047]

[0048] in represents the measured value of the drone state at time k, ε a,k The optimal estimate of is expressed as:

[0049]

[0050] ε ω，k The optimal estimate of is expressed as:

[0051]

[0052] where Ψ(·) represents the function of converting quaternion to Euler angle, ε v，k The optimal estimate of is expressed as:

[0053]

[0054] ε p，k The optimal estimate of is expressed as:

[0055]

[0056] Step 6: Integration of measurement noise is achieved by updating the sensor noise parameters of the Gazebo drone model file, and integration of model bias is achieved by real-time compensation of the flight status data sent by the virtual drone to the flight control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a UAV digital twin model based on parameter identification and correction of the present invention;

[0058] Figure 2 It is the optimal sub-pattern matching error comparison of multiple tasks of the present invention;

[0059] Figure 3 It is a mission example of the present invention, a comparison of flight data fragments in the northeast. DETAILED DESCRIPTION

[0060] The present invention is further described in detail below with reference to the accompanying drawings and embodiments. The present invention proposes a calibration method for a drone cluster digital twin model based on parameter identification, and the specific implementation steps are as follows:

[0061] Step 1: Build a drone digital twin model.

[0062] As attached Figure 1 As shown in the figure, the model includes the following four modules: UAV cluster physical entity, mission control system, flight control system and physical simulation engine; the physical entity represents the actual UAV hardware system, including key sensors such as barometric altimeter, positioning system, three-axis accelerometer and three-axis gyroscope, which are used to collect the status data of the UAV in real time and transmit it to the flight control system to realize the status feedback and control command execution of the UAV; the mission control system is responsible for the high-level command management of the UAV cluster, including mission planning, status update, path planning, waypoint release and parameter identification function modules. The system is dynamically adjusted according to the mission requirements, and generates corresponding path planning and waypoint instructions, and dynamically estimates the measurement noise and model deviation in combination with the measured data; the flight control system uses ArduPilot open source flight control software to convert the instructions of the mission control system into specific control signals, and control the real UAV and the virtual UAV at the same time. This module includes a propulsion system model, a control efficiency model and a flight control model to achieve precise control of the attitude, speed and position of the UAV; the physical simulation engine uses the Gazebo simulation engine to build a virtual model consistent with the real environment, including functions such as environmental interference simulation, rigid body dynamics model and rigid body motion model. By transmitting status data and control instructions, the physical simulation engine can simulate the influence of the external environment, simulate measurement noise, and correct model deviations.

[0063] Step 2: Establish a discrete-time linear space model of the UAV digital twin model.

[0064] The discrete time linear space model of the UAV digital twin model is used to describe the state and control input relationship of the system at different time steps. k-1 represents the system state at time k-1, C k-1 represents the control input at time k-1, M(·) represents the state transfer equation, B(·) represents the control equation, and the ideal state of the system at time k is expressed as:

[0065] X k =M(X k-1 )+B(C k-1 ) (twenty three)

[0066] The actual state of the system at time k is expressed as:

[0067] X k =M(X k-1 )+B(C k-1 )+u k (twenty four)

[0068] Among them, u k Represents the model deviation, and the observed value of the system is expressed as:

[0069] Z k =H(X k )+v k (25)

[0070] Where H(·) represents the measurement equation, v k represents the measurement error, Z k Represents the system observation value at time k; use represents the simulated observation value, represents the simulation state value, and the optimization model of the correction problem is expressed as:

[0071]

[0072] Among them, u opt represents the optimal estimate of the model bias, v opt Represents the optimal estimate of the measurement noise. This discrete-time linear space model can accurately simulate the changes of the system at different time steps by iteratively updating the state of the UAV. In practical applications, the model deviation and measurement noise are continuously corrected by combining the real-time collected data and using parameter identification methods.

[0073] Step 3: Establish the UAV digital twin model measurement model.

[0074] The sensors carried by the drone include a three-axis accelerometer, a three-axis gyroscope, a barometer, and a global positioning system. The three-axis accelerometer measurement model is expressed as:

[0075]

[0076] in B a m Represents the measurement value of the three-axis accelerometer in the body coordinate system, B a represents the true value of the linear acceleration in the body coordinate system B b a Represents the linear acceleration measurement deviation, and its first-order differential is Gaussian white noise n ba , n a Represents a Gaussian white noise matrix with dimension 3×1.

[0077] The three-axis gyroscope measurement model is expressed as:

[0078]

[0079] in: B ω m Represents the measurement value of the three-axis angular velocity meter in the body coordinate system, B ω represents the true value of the angular velocity in the body coordinate system, B b ω Represents the angular velocity measurement deviation, and its first-order differential is Gaussian white noise nbω , n ω Represents a Gaussian white noise matrix with dimension 3×1.

[0080] The barometer and GPS measurement model is expressed as:

[0081]

[0082] in N p m The measured value representing the UAV's NED position, N p represents the true value of the NED position, N b p Represents the position measurement deviation, and its first-order differential is Gaussian white noise n bp ;n p Represents a Gaussian white noise matrix with dimension 3×1.

[0083] Step 4: Establish the state transition model of the drone digital twin model.

[0084] When establishing the state transition model of the UAV digital twin model, the state change of the UAV between consecutive moments can be described by the discretized state transition equation. t It is expressed as:

[0085]

[0086] in, N v represents the three-dimensional velocity of the drone in the north-east coordinate system, q represents the attitude of the drone expressed by quaternion at the current moment, and its leading subscript N and leading superscript B represent the rotation direction of the north-east coordinate system relative to the body coordinate system. The discrete differential state transfer equation of the drone digital twin model is expressed as:

[0087]

[0088] in, represents quaternion multiplication, q{·} represents the conversion function from angular velocity to quaternion, N g represents the gravitational acceleration vector, R represents the rotation matrix, and Δt represents the time interval.

[0089] Step 5: Parameter identification based on the least squares method.

[0090] The sensor measurement error is estimated during the initialization phase before the drone takes off. The state vector of the sensor initialization phase It is expressed as:

[0091]

[0092] A series of measurements during initialization Affected by the additive sensor error Impact:

[0093]

[0094] The mean of the additive error is calculated as:

[0095]

[0096] The variance of the additive error is calculated as:

[0097]

[0098] The model bias is estimated during the flight of the UAV. The discrete equation of the rigid body motion model of the UAV is expressed as:

[0099]

[0100] The state vector of the drone at time k is:

[0101]

[0102] The model deviation of the drone at time k is:

[0103]

[0104] where ε p，k Represented by the three-dimensional position deviation, ε v，k represents the three-dimensional velocity deviation; ε a，k Represents the three-dimensional acceleration deviation, ε ω，k Represents the three-dimensional angular velocity deviation, and the loss function is defined as:

[0105]

[0106]

[0107] in represents the measured value of the drone state at time k, ε a，k The optimal estimate of is expressed as:

[0108]

[0109] ε ω，k The optimal estimate of is expressed as:

[0110]

[0111] where Ψ(·) represents the function of converting quaternion to Euler angle, ε v，k The optimal estimate of is expressed as:

[0112]

[0113] ε p，k The optimal estimate of is expressed as:

[0114]

[0115] A series of model deviations obtained through parameter estimation reflect the differences between the digital twin model and the actual UAV system in dynamic flight.

[0116] Step 6: Integration of measurement noise and model bias.

[0117] The integration of measurement noise is achieved by dynamically adjusting the sensor noise parameters in the drone model file in the Gazebo simulation platform. By modifying the noise configuration of sensors such as accelerometers, gyroscopes, barometers, and GPS, the measurement errors and noise interference encountered by real sensors in the actual environment can be simulated, making the simulation environment closer to reality. The integration of model deviation is achieved by compensating the flight status data of the virtual drone in real time. At each time step, the state output of the virtual drone is adjusted according to the estimated model deviation, and the compensated state data is sent to the flight control system.

[0118] The contents not described in detail in the present application belong to the prior art known to the professional and technical personnel in this field.

Claims

1. A method for calibrating a UAV cluster digital twin model based on parameter identification, characterized in that: Establish a UAV discrete time linear space model, measurement model, and state transition model for digital twins, implement parameter identification based on the least squares method, and correct the position and attitude of the UAV digital twin model. The steps adopted are: Step 1: Establish a UAV digital twin model, which includes four modules: UAV cluster entity, mission control system, flight control system and physical simulation engine. The UAV cluster entity is equipped with sensors to collect flight data and transmit it back to the flight control system and mission control system. The flight control system uses ArduPilot flight control software to control both physical UAVs and virtual UAVs. The mission control system issues mission instructions and waypoint information. The physical simulation engine uses Gazebo to simulate the physical behavior of UAVs in a virtual environment. Step 2: Establish a discrete time linear space model of the UAV digital twin model using X k-1 represents the system state at time k-1, C k-1 represents the control input at time k-1, M(·) represents the state transfer equation, B(·) represents the control equation, and the ideal state of the system at time k is expressed as: X k =M(X k-1 )+B(C k-1 ) (1) The actual state of the system at time k is expressed as: X k =M(X k-1 )+B(C k-1 )+u k (2) Among them, u k Represents the model error, and the observed value of the system is expressed as: Z k =H(X k )+v k (3) Where H(·) represents the measurement equation, v k represents the measurement error, Z k Represents the system observation value at time k; use represents the simulated observation value, represents the simulation state value, and the optimization model of the correction problem is expressed as: Among them, u opt represents the optimal estimate of the model error, v opt represents the optimal estimate of the measurement noise; Step 3: Establish the measurement model of the UAV digital twin model. The three-axis accelerometer measurement model is expressed as: in B a m Represents the measurement value of the three-axis accelerometer in the body coordinate system, B a represents the true value of the linear acceleration in the body coordinate system B b a Represents the linear acceleration measurement deviation, and its first-order differential is Gaussian white noise n ba , n a Represents a Gaussian white noise matrix with a dimension of 3×1; The three-axis gyroscope measurement model is expressed as: in: B ω m Represents the measurement value of the three-axis angular velocity meter in the body coordinate system, B ω represents the true value of the angular velocity in the body coordinate system, B b ω Represents the angular velocity measurement deviation, and its first-order differential is Gaussian white noise n bω , n ω Represents a Gaussian white noise matrix with a dimension of 3×1; The barometer and GPS measurement model is expressed as: in N p m The measured value representing the UAV's NED position, N p represents the true value of the NED position, N b p Represents the position measurement deviation, and its first-order differential is Gaussian white noise n bp ;n p Represents a Gaussian white noise matrix with a dimension of 3×1; Step 4: Establish the state transition model of the UAV digital twin model. The state X of the UAV at a certain moment is expressed as: in, N v represents the three-dimensional velocity of the drone in the north-east coordinate system, q represents the attitude of the drone expressed by quaternion at the current moment, and its leading subscript N and leading superscript B represent the rotation direction of the north-east coordinate system relative to the body coordinate system. The discrete differential state transfer equation of the drone digital twin model is expressed as: in, represents quaternion multiplication, q{·} represents the conversion function from angular velocity to quaternion, N g represents the gravitational acceleration vector, R represents the rotation matrix, and Δt represents the time interval; Step 5: Parameter identification based on the least squares method is used to estimate the sensor measurement error in the initialization phase before the drone takes off. The state vector of the sensor initialization phase It is expressed as: A series of measurements during initialization Affected by the additive sensor error Impact: The mean of the additive error is calculated as: The variance of the additive error is calculated as: The model bias is estimated during the flight of the UAV. The discrete equation of the rigid body motion model of the UAV is expressed as: The state vector of the drone at time k is: The model deviation of the drone at time k is: where ε p，k Represented by the three-dimensional position deviation, ε v，k represents the three-dimensional velocity deviation; ε a，k Represents the three-dimensional acceleration deviation, ε ω，k Represents the three-dimensional angular velocity deviation, and the loss function is defined as: in represents the measured value of the drone state at time k, ε a，k The optimal estimate of is expressed as: ε ω,k The optimal estimate of is expressed as: where Ψ(·) represents the function of converting quaternion to Euler angle, ε v，k The optimal estimate of is expressed as: ε p，k The optimal estimate of is expressed as: Step 6: Integration of measurement noise is achieved by updating the sensor noise parameters of the Gazebo drone model file, and integration of model bias is achieved by real-time compensation of the flight status data sent by the virtual drone to the flight control system.

Images