State estimation method for weak-drive long-endurance gliding state fixed-wing unmanned aerial vehicle

Abstract

The invention relates to the technical field of fixed-wing unmanned aerial vehicle control, in particular to a state estimation method for a fixed-wing unmanned aerial vehicle in a gliding state during weak driving and long endurance. Comprising the steps of sensor data fusion and establishment of an observation model, a judgment process of entering a gliding state, switching of an unmanned aerial vehicle aerodynamic model, establishment of a lightweight particle filter-extended state observer PF-ESO inner and outer ring cascade filtering structure under a weak driving attitude, and working condition feedback control optimization of correlation coefficients. Through a lightweight particle filter-extended state observer cascade structure, the adaptive capacity of the system to airflow disturbance and modeling errors is enhanced, even if airflow change or control delay occurs in long-time gliding, the filter can quickly sense and correct estimation, and long-term state drift is avoided. The robustness of the system to slow disturbance and model mismatch is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fixed-wing UAV control, and particularly to a state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state. Background Art

[0002] In the flight control system of UAVs, especially fixed-wing UAVs, state estimation is a key link for realizing autonomous navigation, attitude control, and path planning. To improve the robustness and estimation accuracy of the system, existing research generally adopts multi-sensor fusion technology and combines filtering algorithms to dynamically estimate the position, speed, attitude, and disturbances of the UAV. Currently, common state estimation algorithms include Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Particle Filter (PF), etc. Among them, the particle filter is widely used in estimation tasks under complex flight states due to its good adaptability to nonlinear systems and non-Gaussian noise. In recent years, the Extended State Observer (ESO) has attracted much attention due to its online compensation ability for system modeling errors and external disturbances, and has also been gradually introduced into the state estimation system, and combined with the particle filter algorithm to form a PF-ESO cascade architecture, further improving the estimation accuracy and robustness.

[0003] However, most existing state estimation systems are established under the assumption of full power or stable cruise flight, and show significant limitations in the application scenarios of UAVs in weakly-driven or even long-endurance gliding states. Specifically, these limitations are particularly prominent in the following aspects:

[0004] 1. Model mismatch in the weakly-driven gliding state: First, the aerodynamic characteristics in the weakly-driven state are significantly different from those in conventional flight. The flight mainly relies on the glide ratio to maintain flight, while the traditional model fails to explicitly introduce the glide ratio parameter, resulting in a systematic deviation in state prediction.

[0005] 2. Insufficient flight state recognition: Common operating condition recognition algorithms cannot accurately identify the weakly-driven gliding state and are prone to misjudging it as an ordinary deceleration or fine-tuning stage. This misjudgment causes the filter to operate under the wrong model assumption, leading to estimation lag and state prediction mismatch.

[0006] 3. Sensitivity to disturbances and response defects: The gliding state is more sensitive to airflow disturbances. The existing ESO parameters are fixed or the gain is too large, which may cause system oscillation or slow response. In such scenarios, EKF and UKF are difficult to cope with the challenge of non-Gaussian disturbances due to the overly simplified noise model assumption, while although PF can theoretically model complex noise, the particle degradation problem is particularly prominent in long-endurance tasks, weakening its robustness.

[0007] 4. Conflict between computational efficiency and energy efficiency: To improve the estimation accuracy, some algorithms increase the number of particles or the filtering frequency in the weak drive stage, which further exacerbates the system's computing power and energy consumption, conflicting with the energy efficiency optimization goal of long-endurance flight missions. Summary of the Invention

[0008] The present invention discloses a state estimation method for a fixed-wing unmanned aerial vehicle in a weak drive long-endurance gliding state. The specific method is as follows:

[0009] Judge whether the unmanned aerial vehicle enters the weak drive long-endurance gliding attitude;

[0010] When the unmanned aerial vehicle is in the weak drive long-endurance gliding attitude, fuse the first multi-source sensor data, combine it with the gliding state dynamics model, perform state estimation on the unmanned aerial vehicle, and obtain the first estimated state;

[0011] Obtain the first observed state, and calculate the first residual by combining the first estimated state and the first observed state with the PF algorithm;

[0012] Input the first residual and the first estimated state into the ESO model, and the ESO model fits the functional relationship between the residual and the estimated state;

[0013] Fuse the second multi-source sensor data, combine it with the gliding state dynamics model, and calculate the second estimated state;

[0014] Input the second estimated state into the ESO model to obtain the second residual corresponding to the second estimated state;

[0015] Correct the second estimated state with the second residual to obtain the final state of the unmanned aerial vehicle.

[0016] Further, to judge whether the unmanned aerial vehicle enters the weak drive long-endurance gliding attitude, the specific method is as follows:

[0017] Calculate the current thrust margin of the unmanned aerial vehicle;

[0018] Compare the calculated current thrust margin with a threshold value;

[0019] When the current thrust margin is greater than the threshold value and the aircraft continuously descends vertically within a preset time period, it is judged that the unmanned aerial vehicle enters the weak drive long-endurance gliding attitude.

[0020] Further, to construct the gliding state dynamics model, the specific method is as follows:

[0021] When it is judged that the unmanned aerial vehicle enters the weak drive long-endurance gliding attitude, switch the conventional dynamic flight model to the gliding state dynamics model; when switching, the readings z = [p n p e p d v nv e v d φ θ ψ], lift coefficient C L and the drag coefficient C D , the weighted average method is used for smooth transition, the formula is as follows:

[0022] f k =(1-α k )f power +α k f glide

[0023] Among them, f power and f glide are the state transfer functions for conventional powered flight and gliding flight, is the transition weight, t0 is the switching start time, and τ is the transition time constant;

[0024] The balance relationship between lift, drag and weight in the flight trajectory coordinate system is:

[0025] F lift =Wcosγ,F drag =Wsinγ,

[0026] Where γ is the flight path angle;

[0027] Lift and drag are obtained by static table lookup:

[0028]

[0029] Where ρ is the air density, S is the wing reference area, and C L (α,q),C D (α,q) are the angle of attack α and the dynamic pressure Changing lookup coefficients;

[0030] The reference lift coefficient C corresponding to the loading altitude band L and the drag coefficient C D , and calculate the initial glide ratio accordingly:

[0031]

[0032] According to the reference lift coefficient C L and the drag coefficient C D , calculate the current lift L0 and drag D0;

[0033] Introducing online correction factors to dynamically adjust aerodynamic parameters C L 、C D ;

[0034] Construct a dynamic model for the gliding state, and the specific formula is as follows:

[0035]

[0036] Among them, the state vector x = [x, y, h, V, γ, ψ] T are respectively the geographical coordinates, barometric altitude, airspeed, path angle and heading angle, and w k is the process noise.

[0037] Furthermore, obtain the first observed state, and the specific method is as follows:

[0038] Use the accelerometer in the IMU to measure the specific force acceleration

[0039] Use the gyroscope in the IMU to measure the angular velocity

[0040] Use the barometer to measure the altitude

[0041] Use GPS to obtain the global position information and speed information

[0042] Perform time alignment on the acquired data;

[0043] Construct the overall observation equation as:

[0044] z = h(x) + v

[0045] In the formula, z is the sensor reading, h(·) is the non-linear observation function, and v is the comprehensive measurement noise.

[0046] Furthermore, calculate the first residual, and the specific method is as follows:

[0047] Fuse the first observed state and update the particle weights. The specific formula is:

[0048]

[0049] The estimated value and the first residual obtained through particle filtering are respectively:

[0050]

[0051] If the number of effective particles is lower than the threshold N th = ηN p , then perform the first residual redistribution.

[0052] Furthermore, fit the functional relationship between the ESO model fitting residual and the estimated state, and the specific method is as follows:

[0053] Bring and r PF,k into the ESO model to construct the extended state:

[0054]

[0055] where d k is the total disturbance, and d k = Bδ k , and L k = diag{l1, l2, …, l n , l d} is the adaptive ESO gain matrix;

[0056] After discretizing the extended state using the Euler method, the state correction is obtained:

[0057]

[0058] where T s represents the sampling period of the Euler method, and the obtained state correction is decomposed:

[0059]

[0060] to obtain the updates of the state and disturbance estimates;

[0061] Extract the first n dimensions from the extended state estimate to obtain the state estimate after ESO correction:

[0062]

[0063] Furthermore, the hyperparameters of the PF algorithm and the ESO model are both dynamically adjusted in real time, and the specific method is as follows:

[0064] Collect the UAV battery state E k , the on-board computing load C k , and the state estimate residual norm ||r PF,k ||2 of the previous cycle in real time;

[0065] Construct a normalized comprehensive performance index, and the specific formula is as follows:

[0066]

[0067] α1 + α2 + α3 = 1

[0068] where α1, α2, and α3 are weight coefficients, and J k ∈[0, 1], the larger it is, the higher the estimation error or energy consumption pressure;

[0069] Dynamically adjust the number of particles N of the PF algorithm according to J k ​p (k) is as follows:

[0070] N p (k) = N min + (N max - N min )(1 - J k )

[0071] where N[[ID=I8]] max , N min are the maximum and minimum number of particles allowed by the algorithm respectively;

[0072] According to J k dynamically adjust the gain matrix L of the ESO k as follows:

[0073] L k = L min + (L max - L min )J k

[0074] where L max , L min are the upper and lower bounds of the ESO gain matrix respectively;

[0075] At the beginning of a new cycle, update the maximum residual record: ||r PF,k || max = max(||r PF,k || max , ||r PF,k ||2).

[0076] Furthermore, correct the second estimated state with the second residual to obtain the final UAV state, and the specific method is as follows:

[0077] Use the second estimated state output by the ESO model in the previous round to calculate the predicted observation through the observation function:

[0078]

[0079] Compare the real-time observation z k with to generate the second residual:

[0080]

[0081] Send the second residual into the ESO model, and according to the ESO gain matrix L k = diag{l1, l2,..., l n , l d} split it into the state gain sub-matrix L x,k = [l1,k , l 2,k , …, l n,k T with the disturbance gain l d,k ;

[0082] Using the second residual to perform linear compensation:

[0083] Δx k = L x,k r ESO,k , Δδ k = l d,k r ESO,k ;

[0084] Obtain the final UAV state and disturbance estimation:

[0085]

[0086] Feed the final state back to the flight control system for aileron and throttle control, and write the latest disturbance estimation back into the ESO model as the prior of the total disturbance for the next cycle.

[0087] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects:

[0088] 1. Through the lightweight particle filter - extended state observer PF - ESO cascade structure, the outer - loop particle filter provides a preliminary state estimation, and the inner - loop extended state observer realizes real - time disturbance estimation and compensation by expanding the disturbance term. This design enhances the system's adaptability to airflow disturbances and modeling errors. Even when encountering airflow changes or control lags during long - term gliding, the filter can quickly sense and correct the estimation, avoiding long - term state drift. This significantly improves the system's robustness to slow - varying disturbances and model mismatches.

[0089] 2. Based on the working condition feedback control, the system can dynamically adjust the number of particles of the particle filter and the gain of the extended state observer, achieving a balance between estimation accuracy and computational load. It can ensure sufficient filtering ability when the disturbance increases and the working conditions are complex, that is, more particles and a faster ESO to maintain the estimation accuracy, and can also reduce the overhead during the stable stage or when resources are limited, that is, reduce the number of particles and lower the sensing frequency to extend the mission time. This adaptive optimization is particularly important in long - endurance missions with limited resources, effectively extending the flight time and reducing energy consumption.

[0090] 3. Through the closed - loop regulation of comprehensive performance indicators, continuously monitor and optimize the estimation accuracy, computational load, and energy consumption to ensure the best performance at all times under different flight stages and mission requirements, thereby enhancing the overall practicality and reliability. ​

[0091] Other advantages, objects, and features of the present invention will be set forth in part in the following description, and in part will be obvious to those skilled in the art based on a review of the following, or can be learned from the practice of the present invention. The objects and other advantages of the present invention can be achieved and obtained by the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The drawings of the present invention are described as follows.

[0093] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0094] Figure 2 It is a schematic diagram of the PS-ESO inner and outer loop cascaded filtering structure model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0095] The present invention will be further described below in conjunction with the drawings and embodiments.

[0096] A state estimation method for a fixed-wing unmanned aerial vehicle in a weakly driven long-endurance gliding state, as Figure 1 shown, the specific steps are as follows:

[0097] S1. Determine whether the unmanned aerial vehicle enters a weakly driven long-endurance gliding attitude. The specific operation is as follows:

[0098] S11. Calculate the thrust margin of the current aircraft. The thrust margin is defined as the ratio of the actual thrust to the required thrust where, T actual is the actual thrust of the aircraft at present, and T required is the thrust required to maintain the current flight state. The thrust margin μ T is used to reflect whether the thrust of the aircraft is sufficient to overcome the resistance and maintain flight.

[0099] S12. Compare the thrust margin with a preset threshold θ thrust to determine whether the aircraft is in a weakly driven gliding state. If μ T < θ thrust lasts for a period of time and the aircraft shows a continuously decreasing vertical speed, it means that the aircraft enters the gliding state and needs to switch to the gliding state processing strategy.

[0100] S13. Once it is determined that the aircraft enters the gliding state, the system will output a gliding state flag to indicate that the filter and the control system should switch to the corresponding gliding state processing strategy. Output working condition flag:

[0101]

[0102] When S glideWhen = 1, the filter will switch to the model adapted to the gliding state, and start the dynamic modeling and estimation process of gliding flight.

[0103] S2. When the UAV is in the gliding attitude with weak drive and long endurance, fuse the first multi-source sensor data, combine it with the gliding state dynamics model, and perform state estimation on the UAV to obtain the first estimated state.

[0104] In this embodiment, when the UAV is in the gliding attitude with weak drive and long endurance, it switches from the conventional dynamics model to the gliding state dynamics model. The specific switching process is as follows:

[0105] S21. When it is determined to enter the gliding state, the model is established as follows:

[0106] The balance relationship of lift, drag and weight in the flight trajectory coordinate system is: F lift =Wcosγ, F drag =Wsinγ, where γ is the flight path angle. The lift and drag are given by the static look-up table method: where ρ is the air density, S is the wing reference area, C L (α,q), C D (α,q) are the look-up table coefficients varying with the angle of attack α and the dynamic pressure respectively.

[0107] S22. Load the C L , C D surfaces corresponding to the height band from the read-only memory and calculate the initial glide ratio accordingly:

[0108] S23. Using the reference lift coefficient C L and the drag coefficient C D obtained in S22, calculate the current lift L0 and drag D0 using the formula in S21.

[0109] S24. Introduce an online correction factor to dynamically adjust the aerodynamic parameters C L , C D , and perform online correction based on the real-time estimation of the aerodynamic database.

[0110] S25. Based on the updated aerodynamic parameters in S24, the system constructs the discrete-time gliding dynamics equation:

[0111]

[0112] where the state vector x = [x, y, h, V, γ, ψ] T , obtained from the observation data in S16, are the geographical coordinates, barometric altitude, airspeed, path angle and heading angle respectively; w kis the process noise. Together with the observation vector obtained by S1 fusion, this glide state equation serves as the prediction model for the PF-ESO cascaded filter, ensuring high-precision estimation of flight attitude and aerodynamic parameters in weakly driven long-endurance missions.

[0113] S26. Use the aerodynamic model constructed in S25 for the state prediction equation. When the weak drive mode is activated, the engine thrust input is no longer considered during state prediction, and instead, the above model is used to calculate the UAV motion parameters.

[0114] S3. Obtain the first observed state, and combine the first estimated state and the first observed state to calculate the first residual using the PF algorithm.

[0115] In this embodiment, as Figure 2 shown, obtaining the first observed state is achieved through sensor data fusion and establishing an observation model. The specific process is as follows:

[0116] S311. Use the accelerometer in the IMU to measure the specific force acceleration which can be approximated as the value of the vehicle acceleration in the body frame minus the gravity: where is the attitude rotation matrix from the body frame to the navigation coordinate system, g is the gravitational acceleration, and n a is the measurement noise. p = [N, E, D] T represents the position vector of the vehicle when using the north-east-down coordinate system.

[0117] S312. Use the gyroscope in the IMU to measure the angular velocity corresponding to the actual rotation rate of the vehicle plus noise. Modeled as: where ω*b is the angular velocity in the body frame, n*ω is the gyroscope noise and drift deviation, and the angular velocity information is fused through the attitude differential relationship

[0118] S313. Use the barometer to measure the altitude and obtain the approximate altitude through atmospheric pressure conversion. The observation model is:

[0119]

[0120] where, is the measured atmospheric pressure, α is the air temperature gradient, and n h is the measurement noise. For low-altitude and small-range flights, a linearized atmospheric model can be used, and the accuracy meets the requirements.

[0121] S314. Use GPS to obtain the global position information and the velocity information The observation model is where np , n v is the measurement noise vector, including the influence of GPS random error and multipath, etc. In high-dynamic scenarios, Doppler frequency shift information can be combined to improve speed estimation.

[0122] S315. To ensure the consistency of timestamps of all sensor data in state estimation, all observations are rounded up to the sampling period Δ according to the nearest timestamp, and missing items are linearly interpolated.

[0123] S316. Through the above sensor observation model, an overall observation equation z = h(x) + v is constructed, where z = [p n p e p d v n v e v d φ θ ψ] contains all sensor readings, h(·) is a non-linear observation function, and v is the comprehensive measurement noise.

[0124] In this embodiment, the first residual is calculated, and the specific steps are as follows:

[0125] S321. At time k, for each particle i = 1, …, N p Sample according to the gliding dynamics prior model where is the process noise, representing the influence of system modeling error and noise, u k-1 is the control input, representing lift and rudder deflection, and f glide (·) represents the gliding dynamics model obtained in S36.

[0126] S322. Use the fused observation vector z k , and update the particle weights:

[0127] S323. The estimated value and residual obtained through particle filtering are respectively:

[0128] S324. If the number of effective particles is lower than the threshold N th = ηN p , perform systematic residual sampling and redistribution to maintain the diversity of the particle set.

[0129] S4. Input the first residual and the first estimated state into the ESO model, and the ESO model fits the functional relationship between the residual and the estimated state.

[0130] In this embodiment, as Figure 2As shown, after fitting the functional relationship, the residual is calculated through the ESO model, and the specific method is as follows:

[0131] S41. Input the obtained and r PF,k into the ESO to construct the extended state:

[0132]

[0133] where d k is the total disturbance, including air flow, actuator lag, etc., and d k =Bδ k , and L k =diag{l1, l2, …, l n , l d} is the adaptive ESO gain matrix.

[0134] S42. Discretize the extended state obtained in S41 using the Euler method to obtain the state correction:

[0135]

[0136] where T s represents the sampling period of the Euler method, thus obtaining the state update equation.

[0137] S43. Extract the first n dimensions from the extended state estimate obtained in S42 to obtain the state estimate after ESO correction:

[0138] S5. Fuse the second multi-source sensor data, combine with the gliding state dynamics model, and calculate the second estimated state.

[0139] S6. Substitute the second estimated state into the ESO model to obtain the second residual corresponding to the second estimated state.

[0140] S7. Correct the second estimated state with the second residual to obtain the final UAV state.

[0141] Correct the second estimated state with the second residual to obtain the final UAV state, and the specific method is as follows:

[0142] Use the second estimated state output by the ESO model in the previous round to calculate the predicted observation through the observation function:

[0143]

[0144] Compare the real-time observation z k with to generate the second residual:

[0145]

[0146] Send the second residual into the ESO model, and according to the ESO gain matrix L k = diag{l1, l2, …, l n , l d}, split it into the state gain sub-matrix L x,k = [l 1,k , l 2,k , …, l n,k T and the disturbance gain l d,k ;

[0147] Use the second residual to make a linear compensation:

[0148] Δx k = L x,k r ESO,k , Δδ k = l d,k r ESO,k ;

[0149] Obtain the final UAV state and disturbance estimation:

[0150]

[0151] Feed the final state back to the flight control system for rudder surface and throttle control, and write the latest disturbance estimation back into the ESO model as the prior of the total disturbance provided in the next cycle.

[0152] In this embodiment, the hyperparameters of the PF algorithm and the ESO model are both dynamically adjusted in real time. The specific method is as follows:

[0153] Collect the state of battery charge E k , the on-board computing load C k and the norm of the state estimation residual ||r PF,k ||2 of the previous cycle in real time, and construct a normalized comprehensive performance index: where α1 + α2 + α3 = 1, which are weight coefficients and can be preset based on the task priority. The larger the index J k ∈ [0, 1], the higher the estimation error or energy consumption pressure. ||r PF,k || max is the historical maximum residual, and C max and E max are the maximum values of the computing load and the state of battery charge respectively. Specifically, the "task priority" can include the following situations:

[0154] ​Long endurance patrol / reconnaissance requires endurance > accuracy > calculation. At this time, the weight combination can be designed as (0.30, 0.20, 0.50);

[0155] Fine mapping requires accuracy > calculation > endurance. At this time, the weight combination can be designed as (0.50, 0.30, 0.20);

[0156] Low-end MCU scenarios or tasks with limited complex calculations require calculation > accuracy ≈ endurance. At this time, the weight combination can be designed as (0.35, 0.45, 0.20);

[0157] During the energy crisis or the return phase, endurance > accuracy is required, and the calculation load should be minimized as much as possible. At this time, the weight combination can be designed as (0.15, 0.15, 0.70).

[0158] According to the proposed comprehensive performance index J k , dynamically adjust the number of particles N of the particle filter PF p (k): N p (k) = N min +(N max -N min )(1 - J k ), where N max , N min are the maximum and minimum number of particles allowed by the algorithm, respectively.

[0159] According to the proposed comprehensive performance index Jk, dynamically adjust the gain matrix L of the extended state observer ESO k : L k = L min +(L max -L min )J k , where L max , L min are the upper and lower bounds of the ESO gain matrix, respectively.

[0160] Write the obtained N p (k) back to update N p (k); write the obtained L k back to update L k .

[0161] At the beginning of a new cycle, update the maximum residual record: ||r PF,k || max = max(||r PF,k || max , ||r PF,k ||2).

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A state estimation method for a fixed-wing unmanned aerial vehicle in a weakly-driven long-endurance gliding state, characterized in that, The specific method is as follows: Determine whether the UAV enters the weakly-driven long-endurance gliding attitude; When the UAV is in the weakly-driven long-endurance gliding attitude, fuse the first multi-source sensor data, combine the gliding state dynamics model, perform state estimation on the UAV, and obtain the first estimated state; Obtain the first observed state, and calculate the first residual by combining the first estimated state and the first observed state using the PF algorithm; Input the first residual and the first estimated state into the ESO model, and the ESO model fits the functional relationship between the residual and the estimated state; Fuse the second multi-source sensor data, combine the gliding state dynamics model, and calculate the second estimated state; Substitute the second estimated state into the ESO model to obtain the second residual corresponding to the second estimated state; Correct the second estimated state with the second residual to obtain the final UAV state.

2. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 1, characterized in that Determine whether the UAV enters the weakly-driven long-endurance gliding attitude. The specific method is as follows: Calculate the current thrust margin of the UAV; Compare the calculated current thrust margin with the threshold; When the current thrust margin is greater than the threshold and the aircraft continuously decreases its vertical height within a preset time period, it is determined that the UAV enters the weakly-driven long-endurance gliding attitude.

3. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 1, characterized in that, Construct the gliding state dynamics model. The specific method is as follows: When determining that the drone enters the gliding attitude during weak-drive long-endurance flight, switch the conventional dynamic flight model to the gliding state dynamic model; during the switch, the readings z = [p n p e p d v n v e v d φ θ ψ] collected by the sensor, the lift coefficient C L and the drag coefficient C D , and use the weighted average method for smooth transition. The formula is as follows: f k =(1 - α k )f power + α k f glide where, f power and f glide are the state transition functions during conventional power flight and gliding flight respectively, are the transition weights, t0 is the switching start time, and τ is the transition time constant; The balance relationship of lift, drag, and weight in the flight trajectory coordinate system is: F lift = W cosγ, F drag = W sinγ, where γ is the flight path angle; The lift and drag are obtained by using the static look-up table method: where ρ is the air density, S is the wing reference area, C L (α,q), C D (α,q) are the look-up table coefficients varying with the angle of attack α and the dynamic pressure respectively; Load the reference lift coefficient C for the corresponding altitude band L and the drag coefficient C D , and calculate the initial glide ratio accordingly: According to the reference lift coefficient C L and the drag coefficient C D , calculate the current lift L0 and drag D0; Introduce an online correction factor to dynamically adjust the aerodynamic parameters C L and C D ; Construct the gliding state dynamics model. The specific formula is: Among them, the state vectors are respectively the geographical coordinates, barometric altitude, airspeed, path angle, and heading angle, and w k is the process noise.

4. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 3, wherein Obtain the first observed state. The specific method is as follows: Measure the specific force acceleration using the accelerometer in the IMU Measure the angular velocity using the gyroscope in the IMU Measuring altitude using a barometer Obtain global position information using GPS and speed information Perform time alignment on the acquired data; Construct the overall observation equation as: z = h(x) + v In the formula, z is the sensor reading, h(·) is the non-linear observation function, and v is the comprehensive measurement noise.

5. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 1, characterized in that Calculate the first residual. The specific method is as follows: Fuse the first observed state and update the particle weights. The specific formula is: The estimated value and the first residual obtained through particle filtering are respectively: If the number of valid particles is lower than the threshold N th = ηN p , then a first residual redistribution is performed.

6. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 5, characterized in that The ESO model fits the functional relationship between the residual and the estimated state. The specific method is as follows: Combine with r PF,k input into the ESO model to construct an extended state: Among them, d k is the total disturbance, d k = Bδ k , L k = diag{l1, l2, …, l n , l d} is the adaptive ESO gain matrix; The extended state is discretized using the Euler method to obtain the state correction: where T s represents the sampling period of the Euler method, and the obtained state correction is decomposed as follows: Obtain the updates of the state and the disturbance estimation; Extract the first n dimensions from the extended state estimation to obtain the state estimation after ESO correction:

7. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 1, characterized in that The hyperparameters of the PF algorithm and the ESO model are both dynamically adjusted in real time. The specific method is as follows: Real-time acquisition of the UAV battery state E k and the on-board computing load C k as well as the state estimation residual norm ||r PF,k ||2; Construct the normalized comprehensive performance index. The specific formula is as follows: α1+α2+α3＝1 where α1, α2, α3 are weight coefficients, and J k ∈ [0, 1], the larger the value, the higher the estimation error or energy consumption pressure; According to J k Dynamically adjust the number of particles N of the PF algorithm p (k), and the specific formula is as follows: N p f(k) = N min + (N max - N min )(1 - J k ) Among them, N max , N min are the maximum and minimum number of particles allowed by the algorithm, respectively; According to J k Dynamically adjust the gain matrix L of the ESO k , and the specific formula is as follows: L k = L min + (L max - L min ) J k where, L max , L min are the upper and lower bounds of the ESO gain matrix, respectively; At the start of a new cycle, update the maximum residual record: ||r PF,k || max = max(||r PF,k || max , ||r PF,k ||2).

8. The state estimation method for a fixed-wing UAV in a weakly-driven long-endurance gliding state according to claim 1, wherein Correct the second estimated state with the second residual to obtain the final UAV state. The specific method is as follows: The second estimated state output by the previous round of the ESO model Calculate the predicted observation through the observation function: Compare the real-time observation z k with to generate a second residual: Send the second residual into the ESO model, and according to the ESO gain matrix L k = diag{l1, l2, …, l n , l d}, split it into the state gain sub-matrix and the disturbance gain l d,k ; Perform linear compensation using the second residual pair : Δx k = L x,k r ESO,k , Δδ k = l d,k r ESO,k ; Obtain the final UAV state and the disturbance estimation: Feed the final state back to the flight control system for control surface and throttle control, and write the latest disturbance estimation back into the ESO model as the prior of the total disturbance for the next cycle.

Images