High-altitude wind field state real-time estimation method and system based on multi-sensor fusion

Abstract

This invention relates to the field of environmental perception and dynamic state estimation technology in high-altitude wind power generation systems, specifically to a method and system for real-time estimation of high-altitude wind field state based on multi-sensor fusion. It simultaneously acquires data from an inertial measurement unit (IMU), a GNSS receiver, and a three-dimensional anemometer. Using the IMU reference point as a benchmark, it employs error-state Kalman filtering for aircraft state estimation and incorporates a long short-term memory (LSTM) network for adaptive adjustment of process noise. Lever arm compensation is achieved through a pre-calibrated installation position vector obtained from offline geometric calibration, eliminating the velocity component induced by aircraft rotation and improving the accuracy of airflow observation. Instantaneous wind speed is then calculated using coordinate transformation and wind trigonometric relationships. This invention improves the robustness of attitude, velocity, and wind speed estimation under aircraft maneuvering or disturbance conditions, providing wind field information for the operation monitoring and strategy optimization of high-altitude wind power generation systems.

Description

Technical Field

[0001] This invention relates to the field of environmental perception and dynamic state estimation technology in high-altitude wind power generation systems, specifically to a method and system for real-time estimation of high-altitude wind field state based on multi-sensor fusion. Background Technology

[0002] In recent years, high-altitude wind power generation has become a highly promising development direction in the renewable energy field due to its significant advantages of stable airflow and high wind energy density at altitudes of 200-800 meters. Unlike traditional ground-based wind power generation, high-altitude wind power generation systems use tethered kites, flying wings, and aerostats as carriers to capture wind energy at high altitudes and convert it into electricity. Its power generation efficiency, operational safety, and control stability are highly dependent on real-time, accurate, and continuous sensing of the wind field conditions at the operating altitude. Key parameters such as wind speed, wind direction, turbulence intensity, and vertical wind shear directly determine the aircraft trajectory planning, tension control, and power generation optimization, and are the core prerequisites for ensuring the efficient and stable operation of the system.

[0003] Current mainstream wind farm measurement technologies are ill-suited to the complex application scenarios of high-altitude wind power, exhibiting significant technical shortcomings. Ground-based wind measurement towers, lidar, and other equipment typically measure at heights below 300 meters, failing to cover the mainstream operating range of high-altitude wind power. Furthermore, fixed measurement points cannot accurately reflect the true characteristics of large-scale, rapidly changing high-altitude wind fields, easily leading to a mismatch between control strategies and actual aerodynamic conditions, thus limiting the improvement of power generation efficiency.

[0004] To achieve in-situ high-altitude wind measurement, current technologies often involve directly mounting sensors such as inertial measurement units and anemometers onto the airborne vehicle. However, during operation, the aircraft undergoes complex motions including pitch, roll, yaw, and linear acceleration, resulting in the sensor output being superimposed with both aircraft motion and airflow disturbance components. Without effective compensation and separation mechanisms, wind speed observations are susceptible to attitude interference, making it impossible to accurately reconstruct the true wind field information and creating a bottleneck in measurement accuracy.

[0005] At the algorithmic level, existing wind field estimation methods mostly employ extended Kalman filtering. This method relies on an accurate system model, making it difficult to adapt to the highly nonlinear and perturbational characteristics of high-altitude wind fields, and it is sensitive to non-Gaussian disturbances such as turbulence. Under conditions such as rapid aircraft maneuvering and severe wind field fluctuations, the filter is prone to divergence and lacks robustness, making it difficult to achieve stable and reliable state estimation.

[0006] In summary, existing technologies have significant shortcomings in terms of measurement coverage, attitude interference resistance, algorithm robustness, and dynamic adaptability. They cannot meet the engineering requirements of high-altitude wind power generation systems for real-time and high-precision estimation of wind field conditions, thus hindering the industry's technological upgrades and large-scale applications. Summary of the Invention

[0007] The purpose of this invention is to propose a real-time estimation method and system for upper-altitude wind field state based on multi-sensor fusion, so as to solve the problems that upper-altitude wind field is difficult to measure directly, wind speed observation is affected by aircraft motion and sensor installation deviation, and estimation robustness is insufficient under high dynamic conditions, so as to achieve stable reconstruction of vertical wind shear profile.

[0008] According to a first aspect of the present disclosure, a real-time estimation method for upper-level wind field state based on multi-sensor fusion is provided, comprising the following steps: Simultaneously acquire the three-axis specific force and three-axis angular velocity output from the inertial measurement unit (IMU) installed on the aircraft, the position and velocity observations output from the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output from the three-dimensional anemometer installed on the outside of the aircraft; Using the installation position of the inertial measurement unit (IMU) as the IMU reference point, an error state Kalman filter (ESKF) is constructed. The triaxial specific force and triaxial angular velocity are used as the time update inputs of the ESKF. A long short-term memory (LSTM) network outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity, dynamically adjusting the process noise covariance matrix. Position and velocity observations are used as the measurement update inputs of the ESKF to correct its state, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. The pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point is obtained, and combined with the three-axis angular velocity after deducting zero bias, the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer is obtained; the induced linear velocity component is removed from the relative airflow velocity to obtain the compensated relative airflow velocity. The compensated relative airflow velocity is converted to the navigation coordinate system based on the calculated attitude, and then vector-synthesized with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

[0009] In one embodiment, the method further includes: reconstructing the vertical wind shear profile of the wind field based on the instantaneous wind speed and its corresponding flight altitude.

[0010] In one embodiment, the states of the error state Kalman filter include: The nominal state is used to describe nonlinear motion, including the position, velocity, and attitude quaternions of the inertial measurement unit reference point in the navigation coordinate system, as well as the accelerometer zero bias and gyroscope zero bias; Error states are used to describe the deviation between the nominal state and the real state, including position error, velocity error, attitude error vector, accelerometer bias error, and gyroscope bias error. The attitude error vector is represented by a three-dimensional rotation vector, which is used to characterize the rotational difference between the nominal attitude quaternion and the true attitude quaternion.

[0011] In one embodiment, the error state of the Kalman filter is corrected, specifically including: The three-axis specific force after subtracting the currently estimated accelerometer zero bias and the three-axis angular velocity after subtracting the currently estimated gyroscope zero bias are used as inputs for nominal state propagation. The kinematic equations of the nominal state are discretized and solved using the median integral method to update the position and velocity of the inertial measurement unit reference point in the navigation coordinate system, and the attitude quaternions are updated using quaternion differential equations. The error state is predicted based on the linearized error state equation, and the measurement is updated using position and velocity observations. The updated error state is then fed back to the nominal state, and the error state is reset to zero.

[0012] In one embodiment, the noise scaling factor used in the output process of a Long Short-Term Memory (LSTM) network specifically includes: The triaxial specific force and triaxial angular velocity data of the inertial measurement unit within the preset time window are processed by a sliding window. Extract statistical features from the data window, wherein the statistical features include at least the mean, standard deviation, and rate of change; The statistical features are input into the long short-term memory network model obtained through offline training, and the process noise scaling factor at the current time step is output. The training samples of the Long Short-Term Memory Network model consist of historical window data from the inertial measurement unit, and the training labels are constructed as reference values ​​for the process noise scaling factor based on the error state Kalman filter measurement residual statistics. According to the process noise scaling factor For the reference process noise covariance matrix Adaptive adjustment is performed to obtain the process noise covariance matrix at the current time: in, Let be the process noise covariance matrix.

[0013] In one embodiment, position and velocity observations are used as measurement update inputs for the error state Kalman filter, specifically: Acquire raw position and ground velocity observations output from the Global Navigation Satellite System (GNSS) receiver and synchronize them in time; When the phase center of the GNSS receiver antenna does not coincide with the reference point of the inertial measurement unit, the original position observation and ground velocity observation are compensated by lever arm based on the pre-calibrated installation position vector between the two and the three-axis angular velocity, and converted into position observation and velocity observation corresponding to the reference point of the inertial measurement unit. The transformed position and velocity measurements are used as the measurement update inputs for the error state Kalman filter.

[0014] In one embodiment, the three-axis angular velocity of the inertial measurement unit after deducting the zero bias of the gyroscope is used to perform a vector cross product operation with the pre-calibrated installation position vector to obtain the linear velocity component introduced at the three-dimensional anemometer measurement center due to the aircraft's rotational motion.

[0015] According to a second aspect of the present disclosure, a real-time upper-level wind field state estimation system based on multi-sensor fusion is provided, comprising: The multi-source sensor data synchronous acquisition module synchronously acquires the three-axis specific force and three-axis angular velocity output by the inertial measurement unit (IMU) installed on the aircraft, the position observation and velocity observation output by the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output by the three-dimensional anemometer installed on the outside of the aircraft. The ESKF state estimation and LSTM adaptive noise adjustment module constructs an error state Kalman filter (ESKF) using the installation position of the inertial measurement unit (IMU) as the IMU reference point. It uses triaxial specific force and triaxial angular velocity as time update inputs to the ESKF, and outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity through a Long Short-Term Memory (LSTM) network to dynamically adjust the process noise covariance matrix. It also uses position and velocity observations as measurement update inputs to correct the state of the ESKF, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. The lever arm compensation and airflow velocity correction module obtains the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, and combines it with the three-axis angular velocity after deducting zero bias to obtain the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer; the induced linear velocity component is removed from the relative airflow velocity to obtain the compensated relative airflow velocity; The instantaneous wind speed calculation module of the navigation coordinate system converts the compensated relative airflow velocity to the navigation coordinate system according to the calculated attitude, and performs vector synthesis with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

[0016] According to a third aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the memory, wherein the processor executes the program to implement the aforementioned method for real-time estimation of upper-level wind field state based on multi-sensor fusion.

[0017] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the aforementioned method for real-time estimation of upper-level wind field state based on multi-sensor fusion.

[0018] The advantages of the above technical solutions adopted in this invention compared with the prior art are as follows: 1. This invention uses the inertial measurement unit reference point as a unified state estimation benchmark, clearly distinguishes the object relationship between the GNSS receiver antenna, the three-dimensional anemometer measurement center, and the pre-calibrated installation position vector, and achieves consistency between the observation and state quantity benchmarks in multi-sensor fusion calculation, effectively reducing the position, velocity, and wind speed calculation errors caused by non-coordinated sensor installation, and improving the overall estimation accuracy.

[0019] 2. By embedding the Long Short-Term Memory Network adaptive noise adjustment mechanism into the error state Kalman filter link, the process noise covariance matrix can be adaptively adjusted based on the temporal statistical characteristics of the historical data of the inertial measurement unit. This can enhance the stability and anti-interference capability of attitude, velocity, position and sensor zero-bias estimation in scenarios of rapid aircraft maneuvering or high-altitude turbulence disturbance.

[0020] 3. By implementing lever arm compensation on the three-dimensional anemometer observations, the local velocity component introduced by the aircraft's rotational motion at the anemometer measurement center can be effectively removed, restoring the relative airflow velocity and significantly improving the physical authenticity and data accuracy of instantaneous wind speed calculation.

[0021] 4. By performing time-domain smoothing and model fitting based on instantaneous wind speed sequences at different altitude levels, the vertical wind shear profile of the high-altitude wind field can be reconstructed in real time and stably, providing comprehensive and reliable environmental data support for the operation monitoring, safety risk assessment, and power generation efficiency optimization strategies of high-altitude wind power generation systems. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0023] Figure 1 This is a schematic diagram of the overall structure of the kite sensor system; Figure 2This is a diagram of a real-time upper-level wind field state estimation architecture based on multi-sensor fusion. Figure 3 This is the flowchart for the Error State Kalman Filter (ESKF). Figure 4 This is a flowchart of the LSTM adaptive noise adjustment mechanism; Figure 5 This is a schematic diagram of the trajectory prediction results; Figure 6 This is a schematic diagram of attitude error; Figure 7 This is a schematic diagram of the upper-level wind field; where (a) is a diagram showing the change of northerly wind speed with altitude, and (b) is a diagram showing the change of easterly wind speed with altitude. Figure 8 These are schematic diagrams of wind speed estimation; (a) is a comparison diagram of northerly wind speed estimation, and (b) is a comparison diagram of easterly wind speed estimation. Figure 9 It is a comparison of wind field estimation and RMSE diagram.

[0024] The numbers in the diagram are explained as follows: 1. Inertial Measurement Unit (IMU); 2. Global Navigation Satellite System (GNSS) Receiver; 3. Barometric Pressure Sensor; 4. Three-dimensional Anemometer. Detailed Implementation

[0025] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0029] like Figure 1 As shown, the invention is illustrated using a high-altitude wind-powered kite as the aircraft platform, but it is not limited to kites. It can also be applied to flying wings, airships, and other high-altitude wind-powered aircraft capable of carrying inertial measurement units, GNSS receivers, and three-dimensional anemometers.

[0030] Example 1: like Figure 2 As shown, this embodiment provides a real-time estimation method for upper-level wind field state based on multi-sensor fusion, including the following steps: S1. Simultaneously acquire the three-axis specific force and three-axis angular velocity output by the inertial measurement unit (IMU) installed on the aircraft, the position and velocity observations output by the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output by the three-dimensional anemometer installed on the outside of the aircraft; Specifically, the inertial measurement unit (IMU) employs high-precision MEMS gyroscopes and accelerometers, and can be installed at the center of the aircraft body, on a rigid top platform, or at other calibrated locations. To unify the state estimation object, the IMU outputs three-axis specific force and three-axis angular velocity in the body coordinate system at a sampling frequency of 100Hz, providing basic data for attitude calculation and high-frequency dynamic estimation.

[0031] The Global Navigation Satellite System (GNSS) receiver employs a multi-frequency receiver supporting RTK. Its antenna can be co-located with the inertial measurement unit (IMU) reference point, or, given a known installation position relationship, deployed on the same rigid structure. The GNSS receiver outputs position and velocity observations, with the velocity observation being the ground velocity in the navigation coordinate system, used to correct for position and velocity drift caused by inertial recursion.

[0032] The three-dimensional anemometer is mounted on the end of a rigid short rod on the outside of the aircraft to avoid direct interference from the wing wake and the complex flow field around the fuselage. The equivalent airflow velocity measurement point of the three-dimensional anemometer is defined as the measurement center of the three-dimensional anemometer, and its output relative airflow velocity is the velocity vector of the airflow relative to this measurement center in the body coordinate system. The pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point is obtained after the system assembly is completed; in an exemplary embodiment, this installation position corresponds to a rigid installation position 0.6 meters in front of the nose of the aircraft. In one embodiment, the pre-calibrated installation position vector is obtained using an offline geometric calibration method. Specifically, after the aircraft is assembled and fixed to the calibration fixture, the installation position of the inertial measurement unit (IMU) is used as the IMU reference point, and a body coordinate system is established with this reference point as the origin. The x-axis of the body coordinate system is set along the nose direction of the aircraft, the y-axis is set along the lateral direction of the aircraft, and the z-axis is determined according to the right-hand coordinate system.

[0033] Three-dimensional coordinate measuring equipment was used to measure the three-dimensional coordinates of the GNSS receiver antenna phase center, the three-dimensional anemometer measurement center, and the inertial measurement unit reference point in the body coordinate system. Let the coordinates of the inertial measurement unit reference point be... The coordinates of the phase center of the GNSS receiver antenna are: The coordinates of the three-dimensional anemometer measurement center are: Where the superscript b represents the body coordinate system. Then the pre-calibration installation position vector of the GNSS receiver antenna relative to the inertial measurement unit reference point is: The pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point is: ,in Lever arm compensation used when converting GNSS receiver observations to inertial measurement unit reference points. Used for calculating the rotation-induced velocity component at the measurement center of a three-dimensional anemometer.

[0034] After calibration, the above and The calibration parameter table is written into the data processing unit. During system operation, the data processing unit directly calls the pre-calibrated installation position vector and performs lever arm compensation by subtracting the gyroscope's zero bias from the three-axis angular velocity of the inertial measurement unit. For the three-dimensional anemometer measurement center, the linear velocity component caused by the aircraft's rotational motion is expressed as... , The three-axis angular velocities of the inertial measurement unit after deducting the gyroscope's zero bias are calculated. The rotational induced velocity component at the three-dimensional anemometer measurement center in the body coordinate system is represented by the symbol ×, which indicates the vector cross product. In one specific embodiment, the three-dimensional anemometer measurement center is located at the end of a rigid short rod approximately 0.6 m in front of the inertial measurement unit reference point. If the x-axis of the body coordinate system points towards the aircraft's nose, its pre-calibrated installation position vector can be represented as... In actual assembly, the pre-calibrated installation position vector is based on the measurement results. To ensure the accuracy of lever arm compensation, the coordinate measurement error between the three-dimensional anemometer measurement center and the GNSS receiver antenna phase center is preferably no greater than 5mm, or no greater than 1% of the corresponding installation arm length.

[0035] The output data from the inertial measurement unit, the GNSS receiver of the global navigation satellite system, and the three-dimensional anemometer are time-stamped, outliers are removed, and calibration parameters are compensated to ensure that all observations enter the fusion solution link under a unified time reference.

[0036] This embodiment uses 100 seconds of simulated flight data, with the flight trajectory employing a variable-frequency "figure-eight" pattern, and the altitude varying between 250m and 350m (e.g., Figure 5 As shown in the figure, the wind shear effect is fully stimulated. The wind field environment adopts a logarithmic wind shear model, with a reference height of 300m, and the wind shear intensities of northerly and easterly winds are 2.0 (m / s) / 100m and 1.5 (m / s) / 100m, respectively.

[0037] S2. Using the installation position of the inertial measurement unit (IMU) as the IMU reference point, an error state Kalman filter (ESKF) is constructed. The triaxial specific force and triaxial angular velocity are used as the time update inputs of the ESKF. A long short-term memory (LSTM) network outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity, dynamically adjusting the process noise covariance matrix. Position and velocity observations are used as the measurement update inputs of the ESKF to correct its state, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. In this embodiment, the reference point of the machine body coordinates corresponding to the installation position of the inertial measurement unit is selected as the reference point of the inertial measurement unit.

[0038] Compared to conventional Kalman filtering, error-state Kalman filtering divides the system state into nominal and error states. The nominal state handles nonlinear dynamics such as attitude and position, while the error state handles linear small disturbances such as position, velocity, attitude, and bias errors. For example... Figure 3 As shown, the error state Kalman filter is explained in detail below: Let the nominal state vector be denoted as Let the error state vector be denoted as ;in, This indicates the position of the IMU reference point in the navigation coordinate system. This indicates the velocity of the IMU reference point in the navigation coordinate system. Represents attitude quaternions, This indicates that the accelerometer has zero bias. This indicates that the gyroscope has zero bias. Indicates positional error. Indicates speed error, This represents the attitude error vector. This indicates the zero bias error of the accelerometer. This indicates the zero bias error of the gyroscope, and the superscript T indicates the transpose of a vector or matrix; , , , , , . ,in , , , , , These are the errors in roll angle, pitch angle, and yaw angle, respectively. , The reason for using a three-dimensional attitude error vector instead of a four-dimensional quaternion error in the error state is that the quaternion of the nominal attitude state satisfies the unit norm constraint, and the actual independent degrees of freedom are 3. Using a three-dimensional small angle vector can describe the small rotational deviation of the nominal attitude relative to the true attitude, and avoids introducing redundant degrees of freedom into the error covariance matrix, thereby improving the stability of filtering.

[0039] During the time update phase, it operates at the inertial measurement unit sampling frequency (100Hz). Let the sampling period at the k-th sampling time be denoted as... The triaxial specific force measured by the inertial measurement unit is denoted as... The triaxial angular velocities measured by the IMU are denoted as... The corrected force after deducting the accelerometer zero bias is denoted as The corrected angular velocity after deducting the gyroscope's zero bias is denoted as... The gravity vector in the navigation coordinate system is denoted as Posture Quaternion The corresponding direction cosine matrix is ​​denoted as The acceleration in the navigation coordinate system is denoted as... Position and speed according to , Update; attitude quaternion by Update, in which The symbol ⊗ represents quaternion multiplication, and Exp(·) represents the exponential mapping of quaternion increments constructed from angular velocity increments.

[0040] In the process of error covariance propagation, the discrete state transition matrix is ​​denoted as... The process noise mapping matrix is The posterior error covariance matrix of the previous time step is The prior error covariance matrix at the current time is The process noise covariance matrix at the current time is The superscript "+" indicates a posterior quantity, and the superscript "-" indicates a prior quantity. The aforementioned The linearization is obtained based on the current attitude, velocity, and zero-bias estimate. The scaling factor output by the LSTM adaptive noise conditioning module is dynamically adjusted.

[0041] During the measurement update phase, the GNSS receiver position and velocity observations are used as the input error state Kalman filter (ESKF), where the velocity observation is the ground speed observation in the navigation coordinate system. If the GNSS receiver antenna and the inertial measurement unit (IMU) reference point are not co-located, the GNSS receiver observations are first converted to the position and velocity observations corresponding to the IMU reference point according to the pre-calibrated installation relationship. The residual vector at the current time is denoted as... The observation vector at the current time is denoted as The observation model is denoted as The prior state is denoted as Where the subscript k represents the k-th filtering time, and the superscript - indicates a priori quantity; then The Kalman gain at the current moment is denoted as... The prior error covariance matrix is ​​denoted as The observation matrix is ​​denoted as The observation noise covariance matrix is ​​denoted as The superscript T denotes matrix transpose, and the superscript -1 denotes matrix inversion; then The prior error state vector at the current moment is denoted as... The posterior error state vector at the current time is denoted as... ;but .

[0042] During the error injection and reset phase, the posterior error state vector is decomposed into... The nominal state is as follows , , , , Inject update, where This represents the small rotation quaternion constructed from the attitude error vector. The covariance update uses the Joseph form, i.e. , where I represents the identity matrix; then the error state is reset to zero to ensure that the next filtering cycle continues to linearize around the new nominal state.

[0043] like Figure 4 As shown, to address the problem that fixed process noise parameters cannot adapt to high-dynamic flight conditions, this invention introduces an adaptive noise adjustment module of LSTM into the error state Kalman filter link. The input to this module is no longer the instantaneous measurement value at a single moment, but a sliding window sequence composed of the IMU triaxial specific force and IMU triaxial angular velocity at the most recent L sampling moments, where L is preferably 10.

[0044] During the feature extraction stage, the mean, standard deviation, and rate of change of each component within the sliding window are calculated in real time to describe the aircraft's maneuverability, disturbance variation trends, and short-term dynamic level. In the gated memory computation stage, the memory cells of the LSTM structure and gating logic are used to maintain short-term historical information, and a process noise scaling factor is output based on the input features. ,in This is a scalar value greater than 0. The scaling factor is related to the reference process noise covariance matrix. The relationship is represented as .

[0045] When the aircraft is detected to be making a figure-eight maneuver, a rapid turn, or other violent maneuvers... Increase this value to appropriately amplify the process noise covariance matrix, thereby improving the filter's response to changes in observations; when the flight state tends to stabilize, Reduce the size to maintain the smoothness of the state estimate and suppress unnecessary fluctuations.

[0046] In one implementation, the LSTM adaptive noise conditioning module obtains network parameters through offline training before deployment and performs only forward inference during system operation. The training dataset consists of numerical simulation data and historical flight data. The numerical simulation data includes inertial measurement unit (IMU) data, GNSS receiver data, three-dimensional anemometer data, flight attitude data, and ground truth wind field data under various conditions such as straight flight, hovering flight, variable-frequency figure-eight flight, rapid turns, climbs, descents, and gust disturbances. The historical flight data includes IMU three-axis specific force, three-axis angular velocity, GNSS receiver position and velocity observations, three-dimensional anemometer relative airflow velocity, and state estimation results obtained after offline smoothing.

[0047] The training samples are constructed using a sliding window approach. Let the input sequence of the inertial measurement unit corresponding to the k-th sampling time be: Where L is the time window length, preferably L=10; the output vector of the inertial measurement unit (IMU) at the k-th sampling time is... ,in This represents the triaxial specific force output by the inertial measurement unit at the k-th sampling time. This represents the three-axis angular velocity output by the inertial measurement unit at the k-th sampling time. Furthermore, the mean, standard deviation, and first-order rate of change of each component in the input sequence can be calculated and used as extended input features to characterize the aircraft's maneuver intensity, disturbance variation trend, and short-term dynamic level within the most recent time window.

[0048] The supervision label for the LSTM adaptive noise conditioning module is a reference value for the process noise scaling factor. In numerical simulation data, since the true position, true velocity, true attitude, and true wind field conditions of the aircraft are known, the reference value can be constructed based on the deviation between the error state Kalman filter prior state and the true state. When the prior state error increases, the increase... When the prior state error is small, reduce The reference value can be constructed from historical flight data based on the measurement residuals and residual covariance during the GNSS receiver measurement update phase. .

[0049] Specifically, let the measurement residual at the k-th measurement update time be... The residual covariance is ,in, Let be the observation vector at the k-th measurement update time. For predicted observations calculated based on prior states, This is the prior state. For the observation matrix, Let be the prior error covariance matrix. To observe the noise covariance matrix, the normalized squared innovation statistic is expressed as: Let the measurement dimension be m, then the reference label for the process noise scaling factor can be constructed as follows: ,in, and These are the lower and upper limits of the process noise scaling factor, respectively. This represents the limiting function. In one implementation, , When the measurement residual is large, Increase, the reference label Increase the size of the reference label to improve the filter's response to rapid maneuvers and wind disturbances; when the measurement residual is small, the reference label... Decrease to maintain the smoothness of state estimation.

[0050] In one embodiment, the LSTM network includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives the triaxial specific force and triaxial angular velocity sequences of the inertial measurement unit (IMU) from the most recent L sampling times, along with statistical features calculated from these sequences. The LSTM layer extracts the temporal variation features of the IMU data within a time window. The fully connected layer maps the hidden state of the LSTM at the last moment to scaling factor features. The output layer outputs a process noise scaling factor. .

[0051] In one embodiment, the LSTM network employs a two-layer LSTM structure, with 32 hidden units in each layer. The output is normalized by a fully connected layer and a sigmoid function, and then mapped to a preset range. Its output relationship is expressed as follows: ,in This represents the hidden state of the LSTM layer at the last moment. and These are the output layer weight matrix and the bias term, respectively. This is the Sigmoid function.

[0052] The training loss function of the LSTM network uses a prediction scaling factor. With reference scaling factor The error between time steps is factored in, and a smoothing constraint on the scaling factor between adjacent time steps is introduced. The loss function can be expressed as: ,in N The number of training samples. The first term is used to constrain the prediction scaling factor to be close to the reference label, and the second term is used to suppress drastic jumps in the scaling factor between adjacent time steps, thereby avoiding unnecessary abrupt changes in the process noise covariance matrix during filtering.

[0053] In one implementation, the training process uses the Adam optimizer with an initial learning rate of 10. -3 The batch size is 64, and the number of training epochs is 100. Training data is divided into training, validation, and test sets. Training is stopped and network parameters are saved when the validation set loss no longer decreases for several consecutive epochs. After training, the LSTM network parameters are stored in the data processing unit.

[0054] During system operation, the data processing unit reads the inertial measurement unit data from the most recent L sampling times in real time using a sliding window method, and performs one LSTM forward inference to obtain the current process noise scaling factor. Subsequently, the data processing unit updates the process noise covariance matrix of the error state Kalman filter according to the following formula. ,in, The reference process noise covariance matrix, This is the process noise covariance matrix actually used at the k-th sampling time. When the aircraft undergoes rapid turns, pitch changes, roll changes, or gust disturbances, the LSTM module outputs a larger value based on the short-time variation characteristics of the inertial measurement unit data. This enhances the filter's ability to track dynamic changes; when the flight state is relatively stable, the LSTM module outputs a smaller value. This is to reduce invalid fluctuations in the state estimation results.

[0055] When the LSTM module output exceeds the preset range, the data processing unit limits it to... Within the range; when the LSTM module experiences communication errors, inference failures, or persistent output anomalies, the system will... Set to 1 and degenerate into the error state Kalman filter mode with a fixed process noise covariance matrix to ensure that the system can still continuously output state estimation results.

[0056] S3. Obtain the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, and combine it with the three-axis angular velocity after deducting zero bias to obtain the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer; remove the induced linear velocity component from the relative airflow velocity to obtain the compensated relative airflow velocity; Because the measurement center of a 3D anemometer is usually not co-located with the reference point of the inertial measurement unit, a local velocity component caused by the rigid body rotation will be generated at the measurement center when the aircraft undergoes angular motion. If this component is not compensated for, the measured relative airflow velocity will include the additional velocity generated by the aircraft's own rotation, thus leading to distortion of the wind field calculation results.

[0057] Therefore, a pre-calibrated installation position vector pointing from the inertial measurement unit reference point to the three-dimensional anemometer measurement center is obtained through an offline geometric calibration method, and this installation position vector is denoted as... The three-axis angular velocity of the inertial measurement unit after deducting the gyroscope's zero bias is denoted as... The linear velocity component introduced by the aircraft's rotation at the center of the three-dimensional anemometer measurement is denoted as... The original anemometer observations are recorded as follows: The compensated relative airflow velocity is denoted as The symbol × represents the cross product of vectors. Then... , .

[0058] S4. The compensated relative airflow velocity is converted to the navigation coordinate system according to the calculated attitude, and vector synthesis is performed with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

[0059] After obtaining the compensated relative airflow velocity, it is transformed from the body coordinate system to the navigation coordinate system based on the aircraft attitude. Let the direction cosine matrix from the body coordinate system to the navigation coordinate system be denoted as... Where the subscript b represents the body coordinate system and the superscript n represents the navigation coordinate system; let the compensated relative airflow velocity vector in the navigation coordinate system be denoted as... The velocity of the IMU reference point in the navigation coordinate system is denoted as... The instantaneous wind speed vector in the navigation coordinate system is denoted as .but Since relative airflow velocity is defined as the velocity vector of the airflow relative to the aircraft, therefore .

[0060] Preferably, the method further includes: reconstructing the vertical wind shear profile of the wind field based on the instantaneous wind speed and its corresponding flight altitude; During the wind shear reconstruction phase, the flight altitude corresponding to the k-th sampling time is denoted as... The instantaneous wind speed vector calculated at the same sampling time is denoted as and establish and A one-to-one correspondence is established, where the subscript k represents the k-th sampling time. The flight altitude can be obtained from barometric altitude (obtained via a barometric pressure sensor), GNSS receiver altitude, or a combination of both.

[0061] Considering the presence of random turbulence and observation noise in high-altitude wind fields, a Kalman smoother is first used to smooth the instantaneous wind speed sequence in the time domain. Then, a logarithmic model or a linear gradient model is used to fit the relationship between wind speed and altitude. When using a logarithmic model, let the wind speed at altitude h be denoted as... Reference height The reference wind speed at the location is recorded as follows: The surface roughness parameter is denoted as ,symbol To represent the natural logarithm, then .

[0062] After wind shear reconstruction, the system can output the average wind speed, wind direction, and wind shear parameters at different altitude levels, and transmit these results to the ground station via the communication and monitoring unit. The ground station can then display the flight trajectory, wind field status, and wind shear variation trends, providing support for operation monitoring, risk assessment, and power generation strategy optimization.

[0063] Taking a set of verification data processing with a sampling duration of 100 seconds as an example, the flight altitude range is 250 meters to 350 meters, the sampling frequency of the inertial measurement unit is 100 Hz, the update frequency of the global navigation satellite system (GNSS) receiver is 10 Hz, and the flight trajectory adopts a variable frequency figure-eight maneuver to fully stimulate the wind shear effect and high dynamic attitude change.

[0064] In the verification data processing, after adopting the method of the present invention, the root mean square error (RMSE) of the position estimation of the inertial measurement unit reference point in the navigation coordinate system is stabilized within 0.342 meters, and the aircraft attitude estimation error is better than 1.5 degrees. In response to the problem of non-center-of-mass installation of the three-dimensional anemometer, the introduction of lever arm effect compensation can significantly reduce the spurious wind speed component introduced by the body angular motion.

[0065] Furthermore, by combining the wind shear model and smoothing filtering, the system can stably characterize the logarithmic trend of wind speed with height. Taking the wind field estimation results as an example, the RMSE of the northward, eastward, and vertical wind field components can be reduced to 0.177 m / s, 0.185 m / s, and 0.097 m / s, respectively, and the overall wind field estimation RMSE can be reduced to 0.274 m / s.

[0066] Depend on Figures 5 to 9 It can be seen that the method of the present invention can not only provide instantaneous estimates of the high-altitude wind field, but also reconstruct its vertical structure stably, thereby providing a basis for the operation monitoring and strategy optimization of the high-altitude wind power generation system.

[0067] The real-time wind field state estimation method of this invention does not depend on the specific shape and structure of the aircraft, but rather on the aircraft's ability to provide a stable sensor installation reference, measurable flight state information, and calibrable sensor relative installation relationships. For tethered kites, flying wings, aerostats, or other high-altitude wind-powered aircraft, as long as they can be equipped with an inertial measurement unit, a GNSS receiver, and a three-dimensional anemometer, and can obtain the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, state estimation, lever arm compensation, instantaneous wind speed calculation, and vertical wind shear reconstruction can be completed according to the process described in this invention.

[0068] For tethered kites, their flight typically involves periodic turns, pitch changes, roll changes, and attitude disturbances caused by the tethering rope. In these situations, the inertial measurement unit (IMU) can be installed in the rigid connection area of ​​the kite body, the GNSS receiver antenna can be positioned on the upper surface of the kite or in other locations with minimal obstruction, and the three-dimensional anemometer can be placed at the end of the rigid short rod on the outer side of the kite to reduce the influence of local flow fields near the wing surface on wind speed measurement. Since tethered kites experience significant angular velocity changes during figure-eight or turning flight, the misalignment error between the three-dimensional anemometer measurement center and the IMU reference point directly affects the wind speed calculation results. Therefore, the local velocity components caused by the aircraft's rotation can be removed according to the lever arm compensation relationship described in this invention.

[0069] For high-altitude wind turbines with flying wings, their flight speeds are typically high, their maneuverability is wide, and their roll, yaw, and pitch velocities vary significantly. In such aircraft, the inertial measurement unit (IMU) can be installed in the rigid section at the center of the wing, the GNSS receiver antenna can be installed in an unobstructed area above the fuselage, and the three-dimensional anemometer can be installed on the leading-edge extension, a short rod at the front of the fuselage, or other locations far from the wing boundary layer and wake influence. Due to the significant variations in flight speed and angular velocity of the flying wing, the velocity of the IMU reference point, the aircraft attitude, and the three-dimensional anemometer installation position vector have a more pronounced impact on the instantaneous wind speed calculation results. Therefore, in the flying wing platform, this invention still uses the IMU reference point as the unified state estimation benchmark, obtains the position, velocity, attitude, and zero-bias estimation results through error state Kalman filtering, and then combines the lever arm compensation results from the three-dimensional anemometer measurement center to calculate the instantaneous wind speed in the navigation coordinate system.

[0070] For aerostat-type high-altitude wind power generation aircraft, their flight speed is typically low and their angular motion amplitude is relatively small, but they are significantly affected by tethered swaying, airflow disturbances, buoyancy changes, and slow altitude changes. In such aircraft, the inertial measurement unit can be installed in the pod, payload compartment, or other rigid connection points, the GNSS receiver antenna can be arranged in the unobstructed upper area, and the three-dimensional anemometer can be arranged away from the disturbance area on the surface of the aerostat via an extended bracket. Due to the low motion frequency of the aerostat, the wind speed sequence smoothing window can be appropriately increased during wind speed estimation to suppress the influence of random turbulence and local swaying on wind shear reconstruction; at the same time, the instantaneous wind speed and vertical wind shear profile can still be obtained using the error state Kalman filtering, process noise adaptive adjustment, lever arm compensation, and wind triangulation process described in this invention.

[0071] The differences between the various aircraft mentioned above mainly lie in the sensor installation location, velocity range, angular velocity variation range, local flow field disturbance region, and wind speed sequence smoothing window selection. These differences do not alter the core processing chain of this invention, namely: using the inertial measurement unit (IMU) reference point as the unified state estimation object; correcting inertial recursion errors using GNSS receiver observations; adaptively adjusting the error state Kalman filter noise covariance matrix using a long short-term memory network; performing lever arm compensation based on the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the IMU reference point; and finally calculating the instantaneous wind speed in the navigation coordinate system based on the compensated relative airflow velocity, aircraft attitude, and IMU reference point velocity.

[0072] Therefore, this invention is applicable to tethered kites, flying wings, aerostats, and other high-altitude wind power generation aircraft capable of carrying inertial measurement units, GNSS receivers, and three-dimensional anemometers. Different aircraft can achieve real-time estimation of high-altitude wind field conditions without altering the overall technical solution of this invention by adjusting parameters such as sensor mounting positions, mounting position vector calibration results, LSTM input window length, upper and lower limits of process noise scaling factors, and wind speed sequence smoothing windows, based on their structural form and aerodynamic characteristics.

[0073] Example 2: This embodiment provides a real-time upper-level wind field state estimation system based on multi-sensor fusion, including: The multi-source sensor data synchronous acquisition module synchronously acquires the three-axis specific force and three-axis angular velocity output by the inertial measurement unit (IMU) installed on the aircraft, the position observation and velocity observation output by the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output by the three-dimensional anemometer installed on the outside of the aircraft. The ESKF state estimation and LSTM adaptive noise adjustment module constructs an error state Kalman filter (ESKF) using the installation position of the inertial measurement unit (IMU) as the IMU reference point. It uses triaxial specific force and triaxial angular velocity as time update inputs to the ESKF, and outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity through a Long Short-Term Memory (LSTM) network to dynamically adjust the process noise covariance matrix. It also uses position and velocity observations as measurement update inputs to correct the state of the ESKF, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. The lever arm compensation and airflow velocity correction module obtains the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, and combines it with the three-axis angular velocity after deducting zero bias to obtain the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer; the induced linear velocity component is removed from the relative airflow velocity to obtain the compensated relative airflow velocity; The instantaneous wind speed calculation module of the navigation coordinate system converts the compensated relative airflow velocity to the navigation coordinate system according to the calculated attitude, and performs vector synthesis with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

[0074] In a preferred embodiment, the system further includes a wind shear reconstruction module. The wind shear reconstruction module is used to perform time-domain smoothing of the instantaneous wind speed sequence based on the instantaneous wind speed in the navigation coordinate system and its corresponding flight altitude, and to fit the wind speed variation with altitude using a logarithmic model or a linear gradient model, further reconstructing the vertical wind shear profile of the wind field.

[0075] The above modules can be deployed on the same device or distributed devices; the division of modules is only a functional logic description and does not limit the specific physical boundaries or implementation order.

[0076] Example 3: An electronic device is provided for running the aforementioned "Real-time Estimation Method for Upper-Altitude Wind Field State Based on Multi-Sensor Fusion". The electronic device includes a processor, a memory, and optional communication interfaces / display devices / input devices, etc.; the memory stores a computer program that can run on the processor, and when the processor executes the program, it implements steps S1 to S4 of the method described in Embodiment 1, specifically including but not limited to: S1. Simultaneously acquire the three-axis specific force and three-axis angular velocity output by the inertial measurement unit (IMU) installed on the aircraft, the position and velocity observations output by the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output by the three-dimensional anemometer installed on the outside of the aircraft; S2. Using the installation position of the inertial measurement unit (IMU) as the IMU reference point, an error state Kalman filter (ESKF) is constructed. The triaxial specific force and triaxial angular velocity are used as the time update inputs of the ESKF. A long short-term memory (LSTM) network outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity, dynamically adjusting the process noise covariance matrix. Position and velocity observations are used as the measurement update inputs of the ESKF to correct its state, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. S3. Obtain the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, and combine it with the three-axis angular velocity after deducting zero bias to obtain the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer; remove the induced linear velocity component from the relative airflow velocity to obtain the compensated relative airflow velocity; S4. The compensated relative airflow velocity is converted to the navigation coordinate system according to the calculated attitude, and vector synthesis is performed with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

[0077] The electronic device hardware can be one of a server, personal computer, workstation, industrial controller, edge computing device, or mobile terminal; the processor can be a general-purpose CPU, GPU, NPU, FPGA, or a combination thereof; the memory can be RAM, ROM, flash memory, or disk array. The device can interact with local / remote data storage (acquiring observation data and outputting inversion results) through a communication interface. The above hardware configuration does not constitute a limitation of the present invention.

[0078] Example 4: A computer-readable storage medium storing a computer program, which, when run on a processor of an electronic device, causes the program to perform the method steps S1 to S4 described in Embodiment 1; the storage medium may be a disk, optical disk, flash memory, solid-state drive, read-only memory, random access memory, or any combination of the above media.

[0079] Those skilled in the art will understand that the modules or steps described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, which can then be stored in a storage device for execution by a computer device. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. This disclosure is not limited to any particular combination of hardware and software.

[0080] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0081] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A real-time estimation method for upper-level wind field state based on multi-sensor fusion, characterized in that, Includes the following steps: Simultaneously acquire the three-axis specific force and three-axis angular velocity output from the inertial measurement unit (IMU) installed on the aircraft, the position and velocity observations output from the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output from the three-dimensional anemometer installed on the outside of the aircraft; Using the installation position of the inertial measurement unit (IMU) as the IMU reference point, an error state Kalman filter (ESKF) is constructed. The triaxial specific force and triaxial angular velocity are used as the time update inputs of the ESKF. A long short-term memory (LSTM) network outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity, dynamically adjusting the process noise covariance matrix. Position and velocity observations are used as the measurement update inputs of the ESKF to correct its state, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. The pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point is obtained, and combined with the three-axis angular velocity after deducting zero bias, the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer is obtained; the induced linear velocity component is removed from the relative airflow velocity to obtain the compensated relative airflow velocity. The compensated relative airflow velocity is converted to the navigation coordinate system based on the calculated attitude, and then vector-synthesized with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

2. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, Also includes: Based on the instantaneous wind speed and its corresponding flight altitude, the instantaneous wind speed is smoothed in the time domain, and the relationship between wind speed and altitude is fitted using a logarithmic model or a linear gradient model to further reconstruct the vertical wind shear profile of the wind field.

3. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, The states of the error state Kalman filter include: The nominal state is used to describe nonlinear motion, including the position, velocity, and attitude quaternions of the inertial measurement unit reference point in the navigation coordinate system, as well as the accelerometer zero bias and gyroscope zero bias; Error states are used to describe the deviation between the nominal state and the true state, including position error, velocity error, attitude error vector, accelerometer bias error, and gyroscope bias error. The attitude error vector is represented by a three-dimensional rotation vector, which is used to characterize the rotational difference between the nominal attitude quaternion and the true attitude quaternion.

4. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, The error state of the Kalman filter is corrected, specifically including: The three-axis specific force after subtracting the currently estimated accelerometer zero bias and the three-axis angular velocity after subtracting the currently estimated gyroscope zero bias are used as inputs for nominal state propagation. The kinematic equations of the nominal state are discretized and solved using the median integral method to update the position and velocity of the inertial measurement unit reference point in the navigation coordinate system, and the attitude quaternions are updated using quaternion differential equations. The error state is predicted based on the linearized error state equation, and the measurement is updated using position and velocity observations. The updated error state is then fed back to the nominal state, and the error state is reset to zero.

5. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, The noise scaling factor used in the output process of the Long Short-Term Memory (LSTM) network specifically includes: The triaxial specific force and triaxial angular velocity data of the inertial measurement unit within the preset time window are processed by a sliding window. Extract statistical features from the data window, wherein the statistical features include at least the mean, standard deviation, and rate of change; The statistical features are input into the long short-term memory network model obtained through offline training, and the process noise scaling factor at the current time step is output. ; According to the process noise scaling factor For the reference process noise covariance matrix Adaptive adjustment is performed to obtain the process noise covariance matrix at the current time: in, Let be the process noise covariance matrix.

6. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, The position and velocity observations are used as the measurement update inputs for the error state Kalman filter, specifically: Acquire raw position and ground velocity observations output from the Global Navigation Satellite System (GNSS) receiver and synchronize them in time; When the phase center of the GNSS receiver antenna does not coincide with the reference point of the inertial measurement unit, the original position observation and ground velocity observation are compensated by lever arm based on the pre-calibrated installation position vector between the two and the three-axis angular velocity, and converted into position observation and velocity observation corresponding to the reference point of the inertial measurement unit. The transformed position and velocity measurements are used as the measurement update inputs for the error state Kalman filter.

7. The real-time estimation method for upper-level wind field state based on multi-sensor fusion according to claim 1, characterized in that, By performing a vector cross product operation between the three-axis angular velocity of the inertial measurement unit after deducting the zero bias of the gyroscope and the pre-calibrated installation position vector, the linear velocity component introduced at the three-dimensional anemometer measurement center due to the aircraft's rotational motion is obtained.

8. A real-time estimation system for upper-level wind field state based on multi-sensor fusion, characterized in that, include: The multi-source sensor data synchronous acquisition module synchronously acquires the three-axis specific force and three-axis angular velocity output by the inertial measurement unit (IMU) installed on the aircraft, the position observation and velocity observation output by the global navigation satellite system (GNSS) receiver, and the relative airflow velocity output by the three-dimensional anemometer installed on the outside of the aircraft. The ESKF state estimation and LSTM adaptive noise adjustment module constructs an error state Kalman filter (ESKF) using the installation position of the inertial measurement unit (IMU) as the IMU reference point. It uses triaxial specific force and triaxial angular velocity as time update inputs to the ESKF, and outputs a process noise scaling factor based on the time-domain characteristics of the triaxial specific force and triaxial angular velocity through a Long Short-Term Memory (LSTM) network to dynamically adjust the process noise covariance matrix. It also uses position and velocity observations as measurement update inputs to correct the state of the ESKF, thereby calculating the position, velocity, attitude of the IMU reference point in the navigation coordinate system, and the zero bias of the IMU. The lever arm compensation and airflow velocity correction module obtains the pre-calibrated installation position vector of the three-dimensional anemometer measurement center relative to the inertial measurement unit reference point, and combines it with the three-axis angular velocity after deducting zero bias to obtain the linear velocity component induced by the aircraft rotational motion at the three-dimensional anemometer; the induced linear velocity component is removed from the relative airflow velocity to obtain the compensated relative airflow velocity; The instantaneous wind speed calculation module of the navigation coordinate system converts the compensated relative airflow velocity to the navigation coordinate system according to the calculated attitude, and performs vector synthesis with the calculated velocity of the inertial measurement unit reference point in the navigation coordinate system to obtain the instantaneous wind speed in the navigation coordinate system.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and running thereon, characterized in that, When the processor executes the program, it implements the real-time estimation method for high-altitude wind field state based on multi-sensor fusion as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the real-time estimation method for upper-level wind field state based on multi-sensor fusion as described in any one of claims 1-7.

Images