A method for intelligent monitoring and predictive guidance of helicopter-towed pods

Abstract

This invention relates to the field of airborne geophysical exploration, and more particularly to an intelligent monitoring and predictive guidance method for helicopter-towed pods. This method involves arranging multiple GPS arrays on the flexible structure of the pod, fusing airborne GPS, radar altimeter, attitude sensor, video, and meteorological data to construct a pod oscillation model incorporating structural modes and identifying deformation parameters online. A multi-model adaptive fusion framework is established, employing extended Kalman filtering and a Bayesian weight update mechanism to achieve state estimation. The spatial wind field gradient is inverted and injected into the forced pendulum dynamics equation to predict the trajectory, calculating the ground contact risk index to trigger graded alarms. Pilot cognitive load is assessed based on eye-tracking and joystick spectral analysis, dynamically simplifying the guidance interface. Lateral / longitudinal oscillation patterns are identified in the body coordinate system, generating differentiated control commands. This invention achieves intelligent monitoring and predictive guidance of the pod's state, improving flight safety and data quality.

Description

Technical Field

[0001] This invention relates to the field of airborne geophysical exploration technology, and in particular to a method for intelligent monitoring and predictive guidance of a helicopter-towed pod. Background Technology

[0002] In helicopter-borne electromagnetic detection operations, the stability of towed pods directly determines data quality and flight safety. Traditional pod monitoring mainly relies on pilot visual observation and single-point GPS positioning. Pilots determine the relative position of the pod through the lower window or voice commands from ground personnel. This method requires a high level of pilot experience during takeoff and landing and cannot monitor dynamic information such as the pod's swing amplitude and altitude in real time. While existing technologies incorporate radar altimetry and inertial sensor-assisted monitoring, they still suffer from three main limitations: First, treating the pod as a rigid body ignores the elastic deformation of the flexible cable and pod structure under airflow disturbances, leading to a significant increase in positioning error with cable length. Second, using fixed-gain filtering or a single physical model for state estimation makes it difficult to adapt to the multimodal characteristics of different flight phases, weather conditions, and payload weights, resulting in model mismatch causing lag or oscillation in state estimation. Third, the human-machine interface uses static information, with fixed display elements for takeoff, operation, and landing. Under high-load conditions, redundant information exacerbates the cognitive burden on pilots, and the lack of an early warning mechanism based on pod dynamics prediction allows for only reactive responses to swaying or ground contact risks, severely restricting the efficiency and safety of airborne geophysical exploration. Therefore, a monitoring and guidance method is urgently needed that can online sense pod flexible deformation, adaptively match multiple dynamic models, predictively assess risks, and intelligently optimize human-machine interaction. Summary of the Invention

[0003] Therefore, the present invention provides an intelligent monitoring and predictive guidance method for helicopter towed pods to solve the aforementioned problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides a method for intelligent monitoring and predictive guidance of a helicopter-towed pod, comprising:

[0005] Step S1: Multiple positioning sensors are arranged on the flexible structure of the pod to collect multi-source heterogeneous data streams from the pod and the helicopter in real time, and simultaneously acquire data streams from the helicopter's onboard GPS, radar altimeter, attitude azimuth instrument and video sensor, as well as meteorological wind field data.

[0006] Step S2: Construct a pod swing model based on the multi-source heterogeneous data stream, use the measurement data from multiple positioning sensors as spatial distribution observation input, and calculate the pod's center of gravity position and deformation parameters online through a parameter identification algorithm;

[0007] Step S3: Use the deformation parameters as input to construct an adaptive fusion framework, and run pod state prediction models with different accuracies in parallel to predict the state and obtain the system state vector.

[0008] Step S4: Based on the system state vector and the meteorological wind field data, the spatial wind field gradient is inverted, the range of the pod's motion trajectory in the future period is predicted through the forced pendulum dynamic equation, and the ground contact risk index is calculated. When the ground contact risk index exceeds the preset safety threshold, a graded alarm is triggered.

[0009] Step S5: Dynamically render the guidance interface on the pilot's display terminal. The guidance interface responds to the pilot's physiological or operational state. When a high-load state is detected, the information presentation is automatically simplified while retaining the core guidance identifiers.

[0010] Step S6: Based on the spatial trajectory characteristics of the pod's relative velocity vector in the airframe coordinate system, a multi-parameter coupled discrimination logic is used to automatically identify the swaying pattern and generate differentiated flight control suppression commands for different swaying patterns.

[0011] Furthermore, the process of step S2 includes:

[0012] The state vector extension of the pod swing model includes the coordinates of the first three structural modes of the pod and the wing surface curvature parameters.

[0013] Using carrier phase differential data from at least three GPS receivers among the multiple positioning sensors as spatial distribution observation inputs, a set of measurement redundancy equations is constructed.

[0014] The online solution is obtained by using an error-weighted regularized least squares parameter identification algorithm. The weight matrix is ​​adaptively and iteratively adjusted according to the product of the squared posterior residuals of each GPS observation and the geometric accuracy factor. The instantaneous center of gravity position of the pod, the modal displacement amplitude and the structural deformation curvature are estimated simultaneously, and spatial continuity constraints are applied to suppress solution divergence.

[0015] Furthermore, the error-weighted regularized least squares parameter identification algorithm further includes: constructing a Tikhonov regularization term to penalize abrupt changes in the second derivative of the modal displacement; the regularization coefficient is dynamically adjusted according to the real-time identified value of the pod cable tension; when the tension is lower than a preset threshold, the regularization intensity is increased to prevent the flexible deformation estimation from overfitting the sensor noise.

[0016] Furthermore, the process of step S3 includes:

[0017] The single pendulum model, the double pendulum model, and the pod swing model are run in parallel, with equal initial weights for all three.

[0018] An adaptive extended Kalman filter is used to perform joint state estimation on the preprocessed multi-source data stream. The normalized residuals and Mahalanobis distances of each model's innovation sequence are calculated. The model fusion weights are dynamically optimized based on a Bayesian posterior probability update mechanism to obtain the system state vector. When the weights of a certain model are lower than 0.05 after 10 consecutive updates, the model is automatically removed.

[0019] Furthermore, the Bayesian post-detection probability update mechanism includes:

[0020] A model switching lag coefficient is introduced to prevent the weights from oscillating violently between adjacent time steps;

[0021] When the posterior probabilities of two or more models are all greater than 0.3, a hybrid model prediction is enabled. The state prediction value is the probability-weighted sum of the predictions of each model, and the covariance cross-fusion term between models is added to quantify the model uncertainty.

[0022] Furthermore, the process of step S4 includes:

[0023] The real-time altitude, vertical velocity, and pitch angular velocity of the pod cable end are extracted from the system state vector.

[0024] By combining radar altimetry data and video sensor data, the terrain height of the projection point below the pod is calculated through spatial interpolation and fusion, and the comprehensive height difference among the three is calculated.

[0025] The comprehensive height difference is normalized by dividing it by the root mean square value of the vertical motion velocity, and a saturation function of the pitch angular velocity is introduced as a dynamic weight reconstruction to build a ground contact risk index.

[0026] Simultaneously, the relative velocity difference between meteorological wind field data and the three GPS points of the pod is used to calculate the local wind field gradient tensor through spatial vector cross product and dot product operations.

[0027] The wind field gradient tensor is injected as a feedforward compensation quantity into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope.

[0028] Furthermore, the process of injecting the wind field gradient tensor as a feedforward compensation quantity into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope includes:

[0029] The wind field gradient tensor is subjected to low-pass filtering to retain gust components with frequencies below 0.5 Hz as effective compensation quantities.

[0030] The filtered wind field gradient is nonlinearly scaled according to the length of the pod cable; the longer the cable, the smaller the compensation gain.

[0031] A compensation time delay matching element is introduced into the dynamic equation of the forced pendulum. The signal transmission delay is calculated based on the relative distance between the helicopter and the pod, and the compensation phase is dynamically adjusted.

[0032] The prediction residuals after compensation are evaluated in real time. If the root mean square value of the residuals for three consecutive prediction periods exceeds the preset mismatch threshold, the compensation gain self-learning adjustment is triggered, and the compensation coefficient is optimized by gradient descent.

[0033] Furthermore, the process of step S5 includes:

[0034] Eye trackers are deployed on pilot helmets or dashboards to monitor physiological parameters in real time, and simultaneously collect joystick displacement signals and perform spectrum analysis.

[0035] Pilot cognitive load assessment results are generated based on physiological state parameters and manipulation spectrum characteristics;

[0036] The complexity of the interface information presentation is dynamically adjusted based on the evaluation results to achieve human-computer interaction that adapts to the load.

[0037] Furthermore, the process of dynamically adjusting the complexity of the interface information presentation based on the evaluation results to achieve adaptive human-computer interaction includes:

[0038] Real-time monitoring of the pilot's gaze point scan rate and pupil diameter change rate is used to calculate the cognitive load index.

[0039] When the cognitive load index exceeds the preset load threshold or the spectrum analysis shows that the energy is concentrated in the 2-5Hz high frequency band and the amplitude is less than 10% of the stroke, it is determined to be a high load stress state. At this time, the complexity of the dynamic adjustment interface information is reduced to a simplified interface configuration that retains only the pod prediction trajectory envelope, direction vector arrows and risk color blocks. The vector arrows use dual encoding of color saturation and geometric length to present the command intensity, and voice broadcasting is disabled to prevent auditory channel overload.

[0040] Furthermore, step S6 includes the following process:

[0041] Construct the phase space trajectory of the relative velocity of the pod in the body coordinate system, and calculate the ratio of its instantaneous velocity components and the magnitude of its angular velocity vector.

[0042] When the lateral sway criterion is met for multiple consecutive sampling periods, a coordinated yaw command and deceleration command are generated.

[0043] When the longitudinal oscillation criterion is met, a speed fine-tuning command is generated;

[0044] The threshold parameter in the criterion is determined in real time by looking up a table based on the length and mass of the pod cable, and the command gain coefficient is adaptively adjusted based on the current flight phase.

[0045] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention uses a distributed GPS array and multi-sensor fusion to transform spatially distributed measurement data into flexible deformation parameters of the gondola using structural mechanics inversion principles, providing accurate initial conditions for the dynamic model; it constructs a multi-model adaptive fusion framework based on Bayesian inference, enabling continuous correction of state prediction errors through innovation feedback, significantly improving state estimation accuracy; it couples hydrodynamic wind field gradient inversion with the forced pendulum dynamic equations, transforming meteorological data into aerodynamic compensation quantities, achieving physical consistency prediction of the gondola trajectory, and giving the ground contact risk index calculation clear mechanical significance; it introduces eye-tracking physiological signals and... By manipulating spectrum analysis and dynamically adjusting information entropy based on the cognitive resource allocation theory of human factors engineering, the pilot's attention is kept focused on core risk factors under high load conditions. Finally, a velocity phase space trajectory is constructed in the airframe coordinate system, and the swaying pattern is identified by multi-parameter coupling discrimination logic. The differentiated generation of yaw-deceleration commands during lateral swaying and speed fine-tuning commands during longitudinal swaying directly corresponds to the pod's dynamic response characteristics, forming a complete physical closed loop from state perception and risk prediction to control commands. This upgrades the traditional passive display to active prediction guidance based on multi-physics coupling, effectively reducing the risk of pod ground contact and swaying amplitude, and improving the quality of geophysical data and flight safety. Attached Figure Description

[0046] Figure 1 A flowchart illustrating an intelligent monitoring and predictive guidance method for a helicopter-towed pod provided by the present invention.

[0047] Figure 2 This is a flowchart illustrating an intelligent monitoring and predictive guidance method for a helicopter-towed pod provided by the present invention. Detailed Implementation

[0048] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0049] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0050] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0051] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0052] Please see Figure 1 As shown, this invention provides a method for intelligent monitoring and predictive guidance of a helicopter-towed pod, comprising:

[0053] Step S1: Multiple positioning sensors are arranged on the flexible structure of the pod to collect multi-source heterogeneous data streams from the pod and the helicopter in real time, and simultaneously acquire data streams from the helicopter's onboard GPS, radar altimeter, attitude azimuth instrument and video sensor, as well as meteorological wind field data.

[0054] Specifically, a distributed sensor array is constructed on the flexible structural frame of the pod. This array consists of no fewer than three high-precision GPS receivers, rigidly mounted at the center of the leading edge, the left side of the trailing edge, and the right side of the trailing edge, respectively. The installation positions must meet the requirements of spatial non-collinear distribution and a spacing of no less than 1 / 3 of the pod length. The sampling frequency of each GPS receiver is uniformly set to 10Hz. The synchronization clock module uses a temperature-controlled crystal oscillator as a reference to generate a PPS pulse signal that is aligned with the GPS second pulse and sends trigger signals to each sensor, ensuring that the data acquisition timestamp synchronization accuracy of the distributed GPS receivers, airborne GPS, radar altimeter, attitude and orientation instrument, and video camera is better than 1ms. The airborne GPS is installed at the helicopter's center of gravity, the radar altimeter is installed on the fuselage belly, the attitude and orientation instrument is fixed near the pod's center of gravity, and the video camera is installed on the helicopter's rear cabin door, with a field of view covering the entire swing range of the pod. Meteorological wind field data is acquired through the helicopter's airborne meteorological radar, with a scanning period set to 5 seconds, acquiring three-dimensional wind field information within a 5-kilometer range in front of the helicopter, and transmitting it to the airborne computing unit via a standard avionics bus.

[0055] Step S2: Construct a pod swing model based on the multi-source heterogeneous data stream, use the measurement data from multiple positioning sensors as spatial distribution observation input, and calculate the pod's center of gravity position and deformation parameters online through a parameter identification algorithm;

[0056] Specifically, step S2 includes the following process:

[0057] The state vector extension of the pod swing model includes the coordinates of the first three structural modes of the pod and the wing surface curvature parameters.

[0058] Specifically, the flexible structure of the pod is modeled using finite element discretization, equating the pod to a spatial Euler-Bernoulli beam. This beam is divided into 8-12 element nodes along its span, with each node retaining 3 translational degrees of freedom and 2 rotational degrees of freedom (torsional degrees of freedom are negligible due to high stiffness). The element mass and stiffness matrices are calculated based on Timoshenko beam theory. After overall assembly, the mass matrix M∈R^(n×n) and stiffness matrix K∈R^(n×n) of the pod structure are obtained. Solving the generalized eigenvalue problem KΦ=ΛM yields the first three modal shape matrices Φ∈R^(n×3) and their corresponding natural frequencies ω1, ω2, ω3. The state vector of the pod's oscillation model is then extended to... The first six terms are the position and velocity of the pod's center of gravity, q1, q2, and q3 are the coordinates of the first three modal modes, and k1 and k2 are the curvature parameters of the pod's leading and trailing edges.

[0059] Using carrier phase differential data from at least three GPS receivers among the multiple positioning sensors as spatial distribution observation inputs, a set of measurement redundancy equations is constructed.

[0060] Specifically, carrier phase differential data from at least three GPS receivers are used as spatial distribution observation inputs to construct a measurement redundancy equation set: for each GPS measurement point location p_i, an observation equation is established. Where Φ_i∈R^(3×3) is the modal shape submatrix of the i-th GPS node, v_i is the observation noise, and all GPS observation equations are stacked to form a redundant observation vector. .

[0061] The online solution is obtained by using an error-weighted regularized least squares parameter identification algorithm. The weight matrix is ​​adaptively and iteratively adjusted according to the product of the squared posterior residuals of each GPS observation and the geometric accuracy factor. The instantaneous center of gravity position of the pod, the modal displacement amplitude and the structural deformation curvature are estimated simultaneously, and spatial continuity constraints are applied to suppress solution divergence.

[0062] Specifically, the error-weighted regularized least squares parameter identification algorithm further includes: constructing a Tikhonov regularization term to penalize abrupt changes in the second derivative of the modal displacement; the regularization coefficient is dynamically adjusted according to the real-time identified value of the pod cable tension; when the tension is lower than a preset threshold, the regularization intensity is increased to prevent the flexible deformation estimation from overfitting the sensor noise.

[0063] Specifically, the weight matrix W = diag(w1, w2, w3) is constructed online using an error-weighted regularized least squares parameter identification algorithm, with its diagonal elements w i The weights represent the confidence weights of the i-th GPS observation. The weight matrix is ​​based on the squared posterior residuals of each GPS observation. With geometric precision factor The product is adaptively iteratively adjusted.

[0064] Specifically, the cost function is defined through online calculation using an error-weighted regularized least squares parameter identification algorithm. The formula for updating the weight matrix is ​​as follows: , These are the adaptive gain coefficients. The initial weights are... The initial weights are set according to the signal-to-noise ratio (SNR) of each GPS receiver; the higher the SNR, the larger the initial weights. Posterior residuals. Based on the current state estimate and observed values The difference is calculated, and the GDOP value is solved in real time by the geometric distribution of GPS satellites.

[0065] Specifically, the regularization coefficient λ is dynamically adjusted based on the real-time identified value of the pod cable tension T, as follows: When the tension T is lower than the preset threshold Tmin, λ increases according to the square law to prevent the flexible deformation estimate from overfitting the sensor noise. The instantaneous center of gravity position, modal displacement amplitude, and structural deformation curvature of the pod are estimated simultaneously, and spatial continuity constraints are applied. ε is the upper limit of deformation set according to the stiffness of the pod material to suppress solution divergence.

[0066] Step S3: Use the deformation parameters as input to construct an adaptive fusion framework, and run pod state prediction models with different accuracies in parallel to predict the state and obtain the system state vector.

[0067] Specifically, step S3 includes the following process:

[0068] The single pendulum model, the double pendulum model, and the pod swing model are run in parallel, with equal initial weights for all three.

[0069] Specifically, during the initialization phase, three pod state prediction models are run in parallel: the single pendulum model simplifies the pod into a mass-cable system, with the state vector containing only three-dimensional position and velocity; the double pendulum model discretizes the cable into two rigid rods, expanding the state vector to 6 dimensions; and the pod swing model adopts the flexible beam coupled dynamics model described in claim 2, with an 11-dimensional state vector. The initial weights of all three are set to μ_k = 1 / 3 (k = 1, 2, 3), forming the model set {M1, M2, M3}.

[0070] An adaptive extended Kalman filter is used to perform joint state estimation on the preprocessed multi-source data stream. The normalized residuals and Mahalanobis distances of each model's innovation sequence are calculated. The model fusion weights are dynamically optimized based on a Bayesian posterior probability update mechanism to obtain the system state vector. When the weights of a certain model are lower than 0.05 after 10 consecutive updates, the model is automatically removed.

[0071] Specifically, each model updates its time independently, that is, based on the state estimate from the previous time step. Covariance Matrix Through the model state transition function Predict the state at that time:

[0072]

[0073] Where u is the helicopter motion input; the covariance prediction is... ,in, Let be the state transition Jacobian matrix. Let be the process noise covariance.

[0074] When multi-source heterogeneous data arrives, calculate the innovation sequence for each model: Among them, Let k be the observation function of model k; the information covariance matrix is: ,in, To observe the Jacobian matrix, To observe the noise covariance, the normalized residual is defined as follows: Its essence is the squared Mahalanobis distance of the innovation vector.

[0075] The formula for updating the posterior probability of each model k is:

[0076]

[0077] Weights are dynamically allocated based on the prediction accuracy (new information size) of each model, with models having smaller prediction errors receiving higher posterior probabilities.

[0078] Specifically, the Bayesian post-detection probability update mechanism includes:

[0079] A model switching lag coefficient is introduced to prevent the weights from oscillating violently between adjacent time steps;

[0080] Specifically, a model switching lag coefficient η∈[0.8,0.95] is introduced to perform a first-order low-pass filter on the weight update:

[0081]

[0082] The weights actually used for fusion are the filtered values. This avoids estimation jitter caused by frequent model switching.

[0083] When the posterior probabilities of two or more models are all greater than 0.3, a hybrid model prediction is enabled. The state prediction value is the probability-weighted sum of the predictions of each model, and the covariance cross-fusion term between models is added to quantify the model uncertainty.

[0084] Specifically, when the posterior probabilities of two or more models are all greater than 0.3, a hybrid model prediction is used, and its state prediction value is the probability-weighted sum of the predictions of each model:

[0085]

[0086] The covariance cross-fusion term is:

[0087]

[0088] The uncertainty differences between models are quantified, improving the conservatism and robustness of the fusion estimation. When the weight of a certain model falls below 0.05 after 10 consecutive updates, the model is automatically removed, its probability is redistributed to the remaining models, and the removal event is logged to help ground maintenance personnel diagnose the cause of model mismatch.

[0089] Step S4: Based on the system state vector and the meteorological wind field data, the spatial wind field gradient is inverted, the range of the pod's motion trajectory in the future period is predicted through the forced pendulum dynamic equation, and the ground contact risk index is calculated. When the ground contact risk index exceeds the preset safety threshold, a graded alarm is triggered.

[0090] Specifically, step S4 includes the following process:

[0091] The real-time altitude, vertical velocity, and pitch angular velocity of the pod cable end are extracted from the system state vector.

[0092] Specifically, extract the real-time vertical height of the end of the pod cable. Vertical motion speed and its root mean square value and pitch angular velocity ;in From the state vector Subtract the cable's geometric sag to obtain the result. Statistical calculations were performed using a sliding time window (5 seconds long).

[0093] By combining radar altimetry data and video sensor data, the terrain height of the projection point below the pod is calculated through spatial interpolation and fusion, and the comprehensive height difference among the three is calculated.

[0094] Specifically, it combines the radar altimetry data obtained in step S1 By combining video sensor data with spatial interpolation data, the terrain height of the projection point below the pod is calculated. Centered on the GPS projection coordinates of the pod, within a 30m×30m ground grid, the mapping between pixel coordinates and ground coordinates is established using the perspective transformation relationship of video images. Bicubic interpolation is then performed on the sparse sampling points of radar altimetry to obtain a high-resolution terrain height field.

[0095] The comprehensive height difference is normalized by dividing it by the root mean square value of the vertical motion velocity, and a saturation function of the pitch angular velocity is introduced as a dynamic weight reconstruction to build a ground contact risk index.

[0096] Specifically, the formula for calculating the ground contact risk index is:

[0097]

[0098] in =0.1 m is a small quantity to prevent division by zero. =5m / s is the saturation constant of the pitch angular velocity; when A yellow alert is triggered when the value is >3.0. When the value is less than 1.5, a red emergency alarm is triggered and an automatic hovering suggestion command is generated.

[0099] Simultaneously, the relative velocity difference between meteorological wind field data and the three GPS points of the pod is used to calculate the local wind field gradient tensor through spatial vector cross product and dot product operations.

[0100] Specifically, the local wind field gradient tensor is calculated using the relative velocity difference between the three GPS points on the pod: Let the relative velocities measured at the three GPS points be... Its spatial location is Then the wind field gradient tensor is:

[0101]

[0102] The wind field gradient tensor is injected as a feedforward compensation quantity into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope.

[0103] Specifically, the process of injecting the wind field gradient tensor as a feedforward compensation into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope includes:

[0104] The wind field gradient tensor is subjected to low-pass filtering to retain gust components with frequencies below 0.5 Hz as effective compensation quantities.

[0105] Specifically, a second-order IIR Butterworth filter is used, with a cutoff frequency set to 0.5Hz, to retain low-frequency gust components and suppress high-frequency turbulence noise; the filter's differential equation is as follows:

[0106]

[0107] Where the coefficient The frequency is determined by the bilinear transform method based on a sampling frequency of 10 Hz and a cutoff frequency of 0.5 Hz.

[0108] The filtered wind field gradient is nonlinearly scaled according to the length of the pod cable; the longer the cable, the smaller the compensation gain.

[0109] Specifically, the filtered wind field gradient is calculated based on the length of the pod cable. Perform nonlinear scaling to compensate for gain: ,in, As the reference gain, =50m is the characteristic length. The longer the cable, the smaller the compensation gain, so as to avoid overcompensation in the case of long cable, which will lead to prediction oscillation.

[0110] A compensation time delay matching element is introduced into the dynamic equation of the forced pendulum. The signal transmission delay is calculated based on the relative distance between the helicopter and the pod, and the compensation phase is dynamically adjusted.

[0111] Specifically, calculating the relative distance between the helicopter and the pod. Signal transmission delay ,in The signal propagation speed within the fiber optic cable (approximately 0.7 times the speed of light); dynamically adjusted compensation phase is... ,in The natural frequency of the pod's oscillation is calculated by multiplying the compensation amount by the phase factor. The forcing term of the equation of the forced pendulum is injected later.

[0112] The prediction residuals after compensation are evaluated in real time. If the root mean square value of the residuals for three consecutive prediction periods exceeds the preset mismatch threshold, the compensation gain self-learning adjustment is triggered, and the compensation coefficient is optimized by gradient descent.

[0113] Specifically, calculating the predicted location Compared with the actual observation location residual vector Its root mean square value is If for three consecutive prediction periods If the preset mismatch threshold emax = 0.5m is exceeded, the compensation gain self-learning adjustment is triggered: the compensation coefficient Kwind is optimized using the gradient descent method, and the update rule is as follows:

[0114]

[0115] Learning rate =0.01, the gradient is approximated by the finite difference method until the residual converges or the maximum number of iterations of 20 is reached.

[0116] Step S5: Dynamically render the guidance interface on the pilot's display terminal. The guidance interface responds to the pilot's physiological or operational state. When a high-load state is detected, the information presentation is automatically simplified while retaining the core guidance identifiers.

[0117] Specifically, step S5 includes the following process:

[0118] Eye trackers are deployed on pilot helmets or dashboards to monitor physiological parameters in real time, and simultaneously collect joystick displacement signals and perform spectrum analysis.

[0119] Specifically, an eye tracker is deployed inside the pilot's helmet visor or in the center above the instrument panel, employing non-contact infrared tracking technology with a sampling frequency of 120Hz and a tracking accuracy better than 0.5°, to acquire the screen coordinates of the gaze point in real time. With pupil diameter The displacement voltage signal of the joystick in the pitch / roll channel is acquired at a frequency of 100Hz via an A / D converter. After calibration and conversion to physical travel percentage, the signal is de-trended and windowed using a Hamming window. A 2048-point FFT is then performed to obtain the spectrum S(f). The energy percentage in the 2-5Hz high-frequency band is then calculated. .

[0120] Pilot cognitive load assessment results are generated based on physiological state parameters and manipulation spectrum characteristics;

[0121] Specifically, the Pilot Cognitive Load Index (CLI) is generated based on physiological state parameters and manipulation spectrum characteristics, and its calculation formula is as follows:

[0122]

[0123] Among them, The fixation point scanning rate was obtained by counting the number of fixation point jumps within 1 second; the weighting coefficients α=0.4, β=0.35, γ=0.25 were fitted and calibrated using data from a pre-conducted pilot physiological experiment. The experiment included high-load take-off and landing missions and low-load cruise missions on a simulator, recording physiological parameters and labeling load levels, and using support vector regression to determine the optimal weights.

[0124] The complexity of the interface information presentation is dynamically adjusted based on the evaluation results to achieve human-computer interaction that adapts to the load.

[0125] Specifically, the process of dynamically adjusting the complexity of the interface information presentation based on the evaluation results to achieve adaptive human-computer interaction includes:

[0126] Real-time monitoring of the pilot's gaze point scan rate and pupil diameter change rate is used to calculate the cognitive load index.

[0127] Specifically, a moving average filter (window length 0.5 seconds) is applied to the pupil diameter to eliminate blinking interference, and the rate of change of pupil diameter is calculated:

[0128]

[0129] in, =1 / 120s.

[0130] When the cognitive load index exceeds the preset load threshold or the spectrum analysis shows that the energy is concentrated in the 2-5Hz high frequency band and the amplitude is less than 10% of the stroke, it is determined to be a high load stress state. At this time, the complexity of the dynamic adjustment interface information is reduced to a simplified interface configuration that retains only the pod prediction trajectory envelope, direction vector arrows and risk color blocks. The vector arrows use dual encoding of color saturation and geometric length to present the command intensity, and voice broadcasting is disabled to prevent auditory channel overload.

[0131] Specifically, when the cognitive load index (CLI) exceeds a preset threshold... =0.75 (normalized value, range 0-1) or high-frequency energy percentage When the value is greater than 0.6 and the joystick amplitude is less than 10% of the travel, it is determined to be a high-load stress state. At this time, the interface adaptive switching logic is triggered: through the OpenGL rendering pipeline, all numerical data (height, speed, coordinates), status indicator lights and mode indicators are gradually hidden within 1 second, leaving only the following three types of core guidance indicators:

[0132] Pod Predicted Trajectory Envelope: The pod's movement range for the next 5 seconds is rendered using a semi-transparent blue cone. The apex of the cone represents the current pod position, and the base radius is determined by the eigenvalues ​​of the prediction covariance matrix.

[0133] Directional Vector Arrow: A green 3D arrow is superimposed in the center of the screen, pointing in the desired manipulation direction. Its color saturation and geometric length are dual-encoded according to the command intensity, and the encoding formula is as follows:

[0134]

[0135] in For the magnitude of the generated manipulation command, The normalization constant is The maximum arrow length is set to 15% of the screen height. When the Ground Contact Index (GSI) > 2.0, a flashing red block is displayed at the edge of the screen, with the flashing frequency proportional to the GSI value. A slight vibration is applied via the torque motor built into the joystick; a vibration frequency of 2Hz indicates a yellow warning, and 5Hz indicates a red warning. A backtrack timer is set; if the CLI remains below a threshold for 5 consecutive seconds, the standard interface is automatically restored. Pilots can also force a switch to any mode by pressing the "Interface Mode" button on the instrument panel.

[0136] Specifically, the dynamic rendering guidance interface further includes an adaptive mode switching function for different flight stages. The system determines the current stage by jointly considering flight altitude and mission status.

[0137] Takeoff Phase Guidance Mode: When the helicopter altitude data acquired in step S1 enters the 20-25m hovering range and remains there for more than 5 seconds, the interface automatically activates the "Alignment Guidance Mode." In this mode, a four-dimensional B-spline guidance curve is generated on the horizontal projection plane from the helicopter's current position to the target hovering point of the pod. The curve color gradually changes along the curve: pure green when the alignment error is less than 0.5m, gradually turning yellowish-green in the 0.5-1.0m range, turning yellow when the error exceeds 1.0m, and turning red with a 2Hz flashing frequency when the error exceeds 2.0m. When the alignment error is less than 0.5m and remains there for more than 3 seconds, a "Allow Descending" text prompt pops up in the center of the interface with a green circular icon as the background, and a voice countdown announcement of the pod's ground clearance begins, with an interval of 1 second.

[0138] Descent Guidance Mode: When the pod's altitude is below 10m, the interface automatically switches to "Careful Descent Monitoring Mode." The vertical altitude display bar is compressed to 0-10m, with the 0-2m range filled in high-saturation red and marked "Danger Zone," the 2-5m range filled in orange and marked "Caution Zone," and the 5-10m range filled in blue and marked "Safe Zone." An overlay of the pod's descent velocity vector arrow is displayed, with the vector originating at the current position of the altitude indicator bar. Its length is proportional to the descent velocity (proportional coefficient 0.5 seconds / meter), and its direction is downward. If the descent velocity vector length exceeds the threshold (corresponding to a velocity of 0.5m / s), a red flashing alarm is triggered, with the flashing frequency proportional to the descent velocity (maximum 10Hz). If the risk index (GSI) calculated in step S4 is simultaneously below 2.0 at this time, the interface additionally displays the mandatory command text "Hover Recommendation."

[0139] Phase determination logic: Flight phase recognition is implemented using a finite state machine. The state transition conditions are jointly determined by the altitude threshold and the pilot's input. The state machine contains three stable states: takeoff phase (altitude < 30m and drop command not activated), operation phase (altitude > 30m and drop completed), and landing phase (altitude < 30m and recovery command activated). The interface mode transitions smoothly during state transitions, with a transition time of 0.5 seconds to avoid visual abrupt changes.

[0140] Step S6: Based on the spatial trajectory characteristics of the pod's relative velocity vector in the airframe coordinate system, a multi-parameter coupled discrimination logic is used to automatically identify the swaying pattern and generate differentiated flight control suppression commands for different swaying patterns.

[0141] Specifically, step S6 includes the following process:

[0142] Construct the phase space trajectory of the relative velocity of the pod in the body coordinate system, and calculate the ratio of its instantaneous velocity components and the magnitude of its angular velocity vector.

[0143] Specifically, the three-dimensional state space coordinates are defined by the positional deviation (Δx, Δy, Δz) of the pod relative to the helicopter, and the corresponding velocity components. This constitutes the phase space velocity field. Relative velocities are extracted from the system state vector. Calculate the instantaneous velocity ratio relationship Where ϵ=0.1m / s is a small amount to prevent division by zero; the pod roll velocity is simultaneously obtained from the attitude azimuth instrument. With pitch angular velocity Calculate the magnitude of the angular velocity vector. .

[0144] When the lateral sway criterion is met for multiple consecutive sampling periods, a coordinated yaw command and deceleration command are generated.

[0145] Specifically, the lateral oscillation criterion is met when five consecutive sampling periods (sampling period 0.1 seconds) satisfy the lateral oscillation criterion. The system is determined to be in a lateral sway mode. At this point, a coordinated yaw command is generated. ,in The yaw gain coefficient is determined by looking up a table: a two-dimensional lookup table array is constructed based on the current pod cable length L_cable (range 10-100m) and pod mass m_pod (range 50-500kg). The values ​​in the table are obtained through offline simulation calibration and interpolation calculation. Vmax=5m / s is the speed saturation limit. A deceleration command is generated synchronously. The deceleration gain Ka is set to 0.15 during takeoff and landing and 0.08 during operation, and is automatically switched according to the flight phase indicator obtained in step S1.

[0146] When the longitudinal oscillation criterion is met, a speed fine-tuning command is generated;

[0147] Specifically, the lateral oscillation criterion is met when five consecutive sampling periods (sampling period 0.1 seconds) satisfy the lateral oscillation criterion. The system is determined to be in longitudinal oscillation mode. A speed fine-tuning command is generated. ,in The speed gain is set to the preset target forward speed (typically 30 m / s during the operational phase). The timescale is 0.2 during takeoff, 0.1 during operation, and 0.3 during landing, adaptively adjusted according to the current flight phase. If the pod is behind the target position (Δx < 0) and If the velocity is less than -0.3 m / s, a slow acceleration command is generated, with the command amplitude adjusted according to... Calculations are performed to ensure smooth adjustment of cable tension.

[0148] The threshold parameter in the criterion is determined in real time by looking up a table based on the length and mass of the pod cable, and the command gain coefficient is adaptively adjusted based on the current flight phase.

[0149] Specifically, the threshold parameter lookup data is stored in the non-volatile memory of the onboard computing unit. The table is a two-dimensional array with dimensions L_cable (10m to 100m, step size 5m) and m_pod (50kg to 500kg, step size 25kg), and each grid cell stores... Three sets of gain values ​​were populated in a table before system installation through wind tunnel testing and flight simulation calibration. During real-time operation, the gain value is obtained through bilinear interpolation based on the current cable length (calculated from the distance between the pod's GPS and the helicopter's GPS) and the pod's mass (either a preset constant or dynamically estimated based on fuel quantity). The command gain is adaptively adjusted based on the current flight phase through a flight phase state machine: the system identifies takeoff, operation, and landing phases based on flight altitude and mission mode, with each phase corresponding to an independent gain scaling factor. The final gain is This ensures optimal suppression performance under different flight missions.

[0150] Specifically, this invention utilizes a distributed GPS array and multi-sensor fusion to transform spatially distributed measurement data into flexible deformation parameters of the gondola using structural mechanics inversion principles, providing precise initial conditions for the dynamic model. Based on Bayesian inference, a multi-model adaptive fusion framework is constructed, allowing state prediction errors to be continuously corrected through innovation feedback, significantly improving state estimation accuracy. By coupling hydrodynamic wind field gradient inversion with the forced pendulum dynamic equations, meteorological data is transformed into aerodynamic compensation quantities, achieving physically consistent prediction of the gondola's trajectory and giving the ground contact risk index calculation clear mechanical significance. Furthermore, eye-tracking physiological signals and manipulation spectrum analysis are introduced. Based on the cognitive resource allocation theory of human factors engineering, information entropy is dynamically adjusted to ensure that the pilot's attention is focused on core risk factors under high load conditions. Finally, a velocity phase space trajectory is constructed in the body coordinate system, and the swaying pattern is identified by multi-parameter coupling discrimination logic. The differential generation of yaw-deceleration commands during lateral swaying and speed fine-tuning commands during longitudinal swaying directly corresponds to the pod's dynamic response characteristics, forming a complete physical closed loop from state perception and risk prediction to control commands. This upgrades the traditional passive display to active prediction guidance based on multi-physics coupling, effectively reducing the risk of pod ground contact and swaying amplitude, and improving the quality of geophysical data and flight safety.

[0151] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0152] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring and predictive guidance of a helicopter-towed pod, characterized in that, include: Step S1: Multiple positioning sensors are arranged on the flexible structure of the pod to collect multi-source heterogeneous data streams from the pod and the helicopter in real time, and simultaneously acquire data streams from the helicopter's onboard GPS, radar altimeter, attitude azimuth instrument and video sensor, as well as meteorological wind field data. Step S2: Construct a pod swing model based on the multi-source heterogeneous data stream, use the measurement data from multiple positioning sensors as spatial distribution observation input, and calculate the pod's center of gravity position and deformation parameters online through a parameter identification algorithm; Step S3: Use the deformation parameters as input to construct an adaptive fusion framework, and run pod state prediction models with different accuracies in parallel to predict the state and obtain the system state vector. Step S4: Based on the system state vector and the meteorological wind field data, the spatial wind field gradient is inverted, the range of the pod's motion trajectory in the future period is predicted through the forced pendulum dynamic equation, and the ground contact risk index is calculated. When the ground contact risk index exceeds the preset safety threshold, a graded alarm is triggered. Step S5: Dynamically render the guidance interface on the pilot's display terminal. The guidance interface responds to the pilot's physiological or operational state. When a high-load state is detected, the information presentation is automatically simplified while retaining the core guidance identifiers. Step S6: Based on the spatial trajectory characteristics of the pod's relative velocity vector in the airframe coordinate system, a multi-parameter coupled discrimination logic is used to automatically identify the swaying pattern and generate differentiated flight control suppression commands for different swaying patterns.

2. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 1, characterized in that, The process of step S2 includes: The state vector extension of the pod swing model includes the coordinates of the first three structural modes of the pod and the wing surface curvature parameters. Using carrier phase differential data from at least three GPS receivers among the multiple positioning sensors as spatial distribution observation inputs, a set of measurement redundancy equations is constructed. The online solution is obtained by using an error-weighted regularized least squares parameter identification algorithm. The weight matrix is ​​adaptively and iteratively adjusted according to the product of the squared posterior residuals of each GPS observation and the geometric accuracy factor. The instantaneous center of gravity position of the pod, the modal displacement amplitude and the structural deformation curvature are estimated simultaneously, and spatial continuity constraints are applied to suppress solution divergence.

3. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 2, characterized in that, The error-weighted regularized least squares parameter identification algorithm further includes: constructing a Tikhonov regularization term to penalize abrupt changes in the second derivative of the modal displacement; the regularization coefficient is dynamically adjusted according to the real-time identified value of the pod cable tension; when the tension is lower than a preset threshold, the regularization intensity is increased to prevent the flexible deformation estimation from overfitting the sensor noise.

4. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 3, characterized in that, The process of step S3 includes: The single pendulum model, the double pendulum model, and the pod swing model are run in parallel, with equal initial weights for all three. An adaptive extended Kalman filter is used to perform joint state estimation on the preprocessed multi-source data stream. The normalized residuals and Mahalanobis distances of each model's innovation sequence are calculated. The model fusion weights are dynamically optimized based on a Bayesian posterior probability update mechanism to obtain the system state vector. When the weights of a certain model are lower than 0.05 after 10 consecutive updates, the model is automatically removed.

5. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 4, characterized in that, The Bayesian post-detection probability update mechanism includes: A model switching lag coefficient is introduced to prevent the weights from oscillating violently between adjacent time steps; When the posterior probabilities of two or more models are all greater than 0.3, a hybrid model prediction is enabled. The state prediction value is the probability-weighted sum of the predictions of each model, and the covariance cross-fusion term between models is added to quantify the model uncertainty.

6. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 5, characterized in that, The process of step S4 includes: The real-time altitude, vertical velocity, and pitch angular velocity of the pod cable end are extracted from the system state vector. By combining radar altimetry data and video sensor data, the terrain height of the projection point below the pod is calculated through spatial interpolation and fusion, and the comprehensive height difference among the three is calculated. The comprehensive height difference is normalized by dividing it by the root mean square value of the vertical motion velocity, and a saturation function of the pitch angular velocity is introduced as a dynamic weight reconstruction to build a ground contact risk index. Simultaneously, the relative velocity difference between meteorological wind field data and the three GPS points of the pod is used to calculate the local wind field gradient tensor through spatial vector cross product and dot product operations. The wind field gradient tensor is injected as a feedforward compensation quantity into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope.

7. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 6, characterized in that, The process of injecting the wind field gradient tensor as a feedforward compensation quantity into the forced vibration term of the forced pendulum dynamics equation to correct the predicted offset direction and amplitude of the pod's motion trajectory envelope includes: The wind field gradient tensor is subjected to low-pass filtering to retain gust components with frequencies below 0.5 Hz as effective compensation quantities. The filtered wind field gradient is nonlinearly scaled according to the length of the pod cable; the longer the cable, the smaller the compensation gain. A compensation time delay matching element is introduced into the dynamic equation of the forced pendulum. The signal transmission delay is calculated based on the relative distance between the helicopter and the pod, and the compensation phase is dynamically adjusted. The prediction residuals after compensation are evaluated in real time. If the root mean square value of the residuals for three consecutive prediction periods exceeds the preset mismatch threshold, the compensation gain self-learning adjustment is triggered, and the compensation coefficient is optimized by gradient descent.

8. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 7, characterized in that, The process of step S5 includes: Eye trackers are deployed on pilot helmets or dashboards to monitor physiological parameters in real time, and simultaneously collect joystick displacement signals and perform spectrum analysis. Pilot cognitive load assessment results are generated based on physiological state parameters and manipulation spectrum characteristics; The complexity of the interface information presentation is dynamically adjusted based on the evaluation results to achieve human-computer interaction that adapts to the load.

9. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 8, characterized in that, The process of dynamically adjusting the complexity of interface information presentation based on the evaluation results to achieve adaptive human-computer interaction includes: Real-time monitoring of the pilot's gaze point scan rate and pupil diameter change rate is used to calculate the cognitive load index. When the cognitive load index exceeds the preset load threshold or the spectrum analysis shows that the energy is concentrated in the 2-5Hz high frequency band and the amplitude is less than 10% of the stroke, it is determined to be a high load stress state. At this time, the complexity of the dynamic adjustment interface information is reduced to a simplified interface configuration that retains only the pod prediction trajectory envelope, direction vector arrows and risk color blocks. The vector arrows use dual encoding of color saturation and geometric length to present the command intensity, and voice broadcasting is disabled to prevent auditory channel overload.

10. The intelligent monitoring and predictive guidance method for a helicopter-towed pod according to claim 9, characterized in that, The process of step S6 includes: Construct the phase space trajectory of the relative velocity of the pod in the body coordinate system, and calculate the ratio of its instantaneous velocity components and the magnitude of its angular velocity vector. When the lateral sway criterion is met for multiple consecutive sampling periods, a coordinated yaw command and deceleration command are generated. When the longitudinal oscillation criterion is met, a speed fine-tuning command is generated; The threshold parameter in the criterion is determined in real time by looking up a table based on the length and mass of the pod cable, and the command gain coefficient is adaptively adjusted based on the current flight phase.

Images