Unmanned aerial vehicle airborne optoelectronic platform multi-band vibration compensation control method based on disturbance observer

Abstract

The present application relates to the technical field of unmanned aerial vehicle photoelectric load vibration control, and particularly relates to a multi-frequency band vibration compensation control method for an unmanned aerial vehicle airborne photoelectric platform based on a disturbance observer, comprising: obtaining unmanned aerial vehicle task phase information, synchronously collecting piezoelectric sensing signals and fiber grating sensing signals, performing time-frequency domain feature fusion and generating a composite feature description in combination with the task phase information; performing mode classification on the current disturbance based on the composite feature description and selecting a corresponding compensation algorithm, predicting the disturbance trend according to the task phase information and activating the compensation algorithm in an event-triggered manner by scheduling computing resources; obtaining an optical axis residual error and a deviation between an actual disturbance and a predicted disturbance, modifying time-frequency domain feature fusion parameters and mode classification discrimination boundaries based on the optical axis residual error, and modifying feature library parameters of the task phase information and a trigger threshold of event-triggered scheduling based on the deviation. The present application realizes adaptive and accurate suppression of wide-frequency band disturbances and improves long-term operation reliability.

Description

Technical Field

[0001] This invention relates to the field of vibration control technology for UAV optoelectronic payloads, and in particular to a multi-band vibration compensation control method for UAV airborne optoelectronic platforms based on a disturbance observer. Background Technology

[0002] UAV-borne optoelectronic platforms (such as optoelectronic pods and EO / IR cameras) are subject to multi-frequency vibration interference from rotors / propellers, engines, aeroelasticity, and the external environment during flight, leading to line-of-sight jitter, image blurring, and target tracking failure. Existing vibration compensation methods mainly combine passive vibration isolation with active control, among which disturbance observers (DOBs) are widely used because they can estimate and suppress disturbances without requiring precise models. However, traditional DOBs are usually designed for a single frequency band or total disturbance, making it difficult to effectively handle multi-source, time-varying vibrations within a wide frequency band of 0.5-500Hz simultaneously; moreover, limited onboard computing resources prevent the simultaneous execution of multiple complex DOB algorithms. In addition, existing methods are mostly based on real-time feedback, lacking forward-looking prediction of disturbance trends, and are unable to cope with sudden disturbance changes caused by mission phase transitions. Summary of the Invention

[0003] To this end, the present invention provides a multi-band vibration compensation control method for an airborne optoelectronic platform of an unmanned aerial vehicle based on a disturbance observer. By integrating mission phase prediction and multi-physics field perception, it realizes adaptive identification of disturbance patterns and event-triggered scheduling of computing resources, and uses a dual feedback closed loop to correct key parameters, so as to achieve efficient suppression of wide-band vibration under limited computing resources.

[0004] To achieve the above objectives, the present invention provides a multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer, comprising: The mission phase information of the UAV is obtained, and piezoelectric sensing signals and fiber optic grating sensing signals at key locations of the optoelectronic platform structure are collected simultaneously. The time-frequency domain features of the piezoelectric sensing signals and fiber optic grating sensing signals are fused, and a composite feature description is generated by combining the mission phase information. The collaborative scheduling based on the composite feature description includes: classifying the current disturbance into a pattern and selecting the corresponding compensation algorithm based on the classification result; predicting the future disturbance trend in the time domain based on the task stage information and scheduling computing resources in an event-triggered manner and activating the compensation algorithm. Obtain the compensated optical axis residual error, obtain the deviation between the actual disturbance and the predicted disturbance, correct the parameters of the time-frequency domain feature fusion and the discrimination boundary of the pattern classification based on the optical axis residual error, and correct the feature library parameters of the task stage information and the trigger threshold of the event-triggered scheduling based on the deviation between the actual disturbance and the predicted disturbance.

[0005] As a preferred technical solution for multi-band vibration compensation control of UAV airborne optoelectronic platforms based on disturbance observers, the process of generating composite feature descriptions includes: The flight control system obtains the mission phases of the UAV in real time, including the hovering phase, forward flight phase, turning phase, and carrier landing glide phase. Based on the task stage, the dominant frequency bands and energy levels of the corresponding stage are extracted from the preset disturbance spectrum feature library to generate a disturbance trend prediction signal; The high-frequency stress wave signal sensed by the piezoelectric sensor and the low-frequency deformation signal sensed by the fiber optic grating sensor are synchronously acquired through a shared signal processing circuit. High-frequency features and low-frequency features are extracted separately, and the high-frequency features and low-frequency features are weighted and fused to obtain a real-time sensing feature vector. Align the disturbance trend prediction signal with the real-time sensing feature vector in the time dimension, and determine the aligned composite data as the composite feature description.

[0006] As a preferred technical solution for the multi-band vibration compensation control method of UAV airborne optoelectronic platform based on disturbance observer, the parameters of time-frequency domain feature fusion include the fusion weighting coefficient of piezoelectric sensing signal and fiber optic grating sensing signal, the filter cutoff frequency during high-frequency feature extraction, and the filter cutoff frequency during low-frequency feature extraction.

[0007] As a preferred technical solution for the multi-band vibration compensation control method of UAV airborne optoelectronic platform based on disturbance observer, the composite feature description is input into a pre-stored one-dimensional convolutional neural network to classify the current disturbance and output the classification result; The classification results include low-frequency dominant mode, mid-frequency dominant mode, high-frequency dominant mode, and multi-frequency mixed mode. Specifically, the low-frequency dominant mode corresponds to an energy proportion in the 0.5-5Hz frequency band exceeding a first threshold, the mid-frequency dominant mode corresponds to an energy proportion in the 5-50Hz frequency band exceeding a second threshold, the high-frequency dominant mode corresponds to an energy proportion in the 50-500Hz frequency band exceeding a third threshold, and the multi-frequency hybrid mode corresponds to an energy proportion in at least two frequency bands exceeding their respective thresholds.

[0008] As a preferred technical solution for the multi-band vibration compensation control method of UAV airborne optoelectronic platform based on disturbance observer, the corresponding compensation algorithm is selected according to the classification results, including: If the classification result is a low-frequency dominant mode, the sliding mode perturbation observer is selected as the compensation algorithm. If the classification result is a mid-frequency dominant mode, the improved repetitive control disturbance observer is selected as the compensation algorithm; If the classification result is a high-frequency dominant mode, the piezoelectric active actuator feedforward compensation algorithm is selected as the compensation algorithm. If the classification result is a multi-frequency mixed mode, a weighted parallel disturbance observer is selected as the compensation algorithm. The weighted parallel disturbance observer is composed of a sliding mode disturbance observer, an improved repetitive control disturbance observer, and a piezoelectric active actuator feedforward compensation algorithm connected in parallel according to preset weights.

[0009] As a preferred technical solution for a multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer, the method predicts future time-domain disturbance trends based on the mission phase information to determine the event triggering mode, and schedules computing resources and activates the compensation algorithm according to the event triggering mode, including: If it is predicted that a sudden change in the frequency band of the disturbance or the energy level of the disturbance will exceed the preset energy threshold in the future time domain, a pre-activation event is generated, and in response to the generation of the pre-activation event, the corresponding compensation algorithm is switched from the low-power standby state to the running state before the disturbance arrives, and the corresponding computing resources are allocated. If the predicted disturbance energy level in the future time domain is lower than a preset calm threshold, a low-power event is generated, and in response to the low-power event, the current compensation algorithm is switched to the basic disturbance observer, and excess computing resources are released. If the deviation between the real-time perceived feature vector and the disturbance trend prediction signal exceeds a preset deviation threshold, a rescheduling event is immediately generated, and in response to the rescheduling event, the compensation algorithm is re-executed to select and allocate computational resources.

[0010] As a preferred technical solution for the multi-band vibration compensation control method of UAV airborne optoelectronic platform based on disturbance observer, the basic disturbance observer is a third-order linear disturbance observer. The third-order linear disturbance observer runs continuously in low-power mode, and its computational resource utilization rate is lower than that of any of the sliding mode disturbance observer, improved repetitive control disturbance observer, and weighted parallel disturbance observer.

[0011] As a preferred technical solution for the multi-band vibration compensation control method of an UAV airborne optoelectronic platform based on a disturbance observer, the process of correcting the parameters of the time-frequency domain feature fusion and the discrimination boundary of the mode classification based on the optical axis residual error includes: The power spectral density of the optical axis residual error after compensation is calculated in real time, and the residual energy values ​​of the low-frequency band, mid-frequency band and high-frequency band are determined respectively. The parameters of the time-frequency domain feature fusion are corrected based on the residual energy values ​​of the low-frequency, mid-frequency, and high-frequency bands, wherein: If the residual energy value in the low-frequency band is greater than the low-frequency threshold, then increase the fusion weight coefficient of the fiber grating sensing signal in the time-frequency domain feature fusion, and adjust the filter cutoff frequency during low-frequency feature extraction to shift towards the low-frequency direction. If the residual energy value in the mid-frequency band is greater than the mid-frequency threshold, then the discrimination boundary of the corresponding mid-frequency dominant mode in the one-dimensional convolutional neural network is adjusted to shift towards the direction of decreasing mid-frequency energy. If the residual energy value in the high-frequency band is greater than the high-frequency threshold, then increase the fusion weight coefficient of the piezoelectric sensing signal in the time-frequency domain feature fusion, and adjust the filter cutoff frequency during high-frequency feature extraction to shift towards the high-frequency direction.

[0012] As a preferred technical solution for the multi-band vibration compensation control method of UAV airborne optoelectronic platform based on disturbance observer, the process of correcting the feature library parameters of the task stage information and the trigger threshold of event-triggered scheduling based on the deviation between the actual disturbance and the predicted disturbance includes: Obtain the spectral distribution of the actual disturbance and calculate the deviation vector between the actual disturbance and the disturbance trend prediction signal in the frequency domain; The deviation vector is used as a correction value to update the statistical parameters of the dominant frequency band and energy level for the corresponding task stage in the disturbance spectrum feature library. The deviation between the actual triggering time and the ideal triggering time of the pre-activation event is monitored in real time. If the deviation exceeds the time deviation threshold, the energy threshold is adjusted to correct the prediction sensitivity. The triggering frequency of rescheduling events is monitored in real time. If the triggering frequency exceeds the frequency threshold, the preset deviation threshold is adjusted to correct the error in the direction of increasing tolerance.

[0013] As a preferred technical solution for multi-band vibration compensation control of UAV airborne optoelectronic platforms based on disturbance observers, it also includes a collaborative optimization step: The correction amount based on the optical axis residual error and the correction amount based on the deviation between the actual disturbance and the predicted disturbance are input into the collaborative optimization module; The collaborative optimization module adopts multi-objective Bayesian optimization, with the optimization objectives being to minimize the optical axis residual error and the computational resource utilization rate. At the same time, it determines the global optimal match of time-frequency domain feature fusion parameters, pattern classification and discrimination boundaries, feature library parameters, and event triggering thresholds. The globally optimal match is used as the initial parameter for the next control cycle.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: The multi-band vibration compensation control method for UAV airborne optoelectronic platforms based on disturbance observers provided by this invention organically integrates task-phase feedforward prediction, piezoelectric-fiber multiphysics field fusion sensing, event-triggered resource scheduling, and dual feedback correction of compensation effect and prediction deviation. This achieves adaptive and precise suppression of wide-band disturbances of 0.5-500Hz under the condition of limited airborne computing resources. This invention does not rely on precise mathematical models and can dynamically adapt to task phase switching, disturbance frequency drift, and environmental changes, significantly improving the line-of-sight stability accuracy and long-term operational reliability of the optoelectronic platform, while reducing the average power consumption of the system by about 30%-40%. This provides key technical support for tasks such as autonomous landing of shipborne UAVs and reconnaissance in complex environments. In particular, this invention solves the technical problem of traditional disturbance observers relying solely on real-time feedback and lacking foresight by setting up a composite feature description generation mechanism that combines mission phase prediction with piezoelectric-fiber fusion sensing. This invention acquires mission phase information from the flight control system in real time and generates a disturbance trend prediction signal by combining it with a pre-set disturbance spectrum feature library. Simultaneously, it uses piezoelectric sensors to sense high-frequency stress waves and fiber optic grating sensors to sense low-frequency deformations. These are then synchronously acquired and fused in the time-frequency domain through a shared signal processing circuit to obtain a real-time sensing feature vector. The predicted signal and the real-time feature vector are aligned in the time dimension to form a composite feature description. This setup enables the system to possess both foresight and accuracy. On the one hand, mission phase-based prediction allows the system to anticipate upcoming disturbance feature changes, reserving response time for subsequent resource scheduling. On the other hand, the physical complementarity of piezoelectric and fiber optic fusion overcomes the limitation of a single sensor not being able to simultaneously cover both low-frequency and high-frequency bands, ensuring high information integrity of the composite feature description across the entire 0.5-500Hz frequency band. After spatiotemporal alignment, this provides a richer and more reliable decision-making basis for pattern classification and event scheduling than a single data source. In particular, this invention solves the computational resource conflict of multi-band compensation algorithms not being able to run simultaneously on embedded platforms by setting up an event-triggered collaborative scheduling mechanism: Based on composite feature description, this invention uses a one-dimensional convolutional neural network to classify disturbances in real time into four modes: low-frequency dominant, mid-frequency dominant, high-frequency dominant, and multi-frequency hybrid, and maps them to sliding mode disturbance observers, improved repetitive control disturbance observers, piezoelectric active actuator feedforward compensation, and weighted parallel disturbance observers, respectively; at the same time, it generates pre-activation events, low-power events, and rescheduling events based on the prediction results of the task stage, dynamically activating the corresponding compensation algorithms and allocating computational resources; this setting realizes resource management of on-demand activation and prediction first, that is, the pre-activation event ensures that the algorithm switch is completed before the disturbance changes, eliminating response delay; the low-power event switches the system to the basic linear disturbance observer during the disturbance calm period, greatly reducing energy consumption; the rescheduling event serves as a safety redundancy, forcing a re-decision when the prediction deviation exceeds the limit; compared with the traditional method of always running a single complex algorithm or switching at a fixed period, this invention achieves the unity of multi-band fine compensation and low-power operation under the same hardware conditions; In particular, this invention solves the performance degradation problem caused by operating condition drift and sensor aging in open-loop or single-feedback systems during long-term operation by setting up a dual-feedback correction and collaborative optimization module. This invention constructs two independent feedback branches: the first branch, based on the compensated optical axis residual error power spectral density, corrects the piezoelectric-fiber fusion weight coefficients, filter cutoff frequency, and mode classification discrimination boundary, so that the perception layer and decision layer are continuously optimized with the compensation effect; the second branch, based on the frequency domain deviation vector between the actual disturbance and the predicted disturbance, corrects the statistical parameters of the feature library of the task stage and the energy threshold and deviation threshold of the event triggering scheduling. The value enables the prediction model and scheduling strategy to dynamically match the actual working conditions. This dual feedback branch inputs the correction amount to the multi-objective Bayesian collaborative optimization module, aiming to minimize the optical axis residual error and the computational resource utilization rate, and determines the globally optimal matching parameters online. This setting enables the system to have adaptive evolution capabilities. The compensation effect feedback forms an inner closed loop of perception, classification, compensation, and correction of perception, while the prediction deviation feedback forms an outer closed loop of prediction, scheduling, evaluation, and correction of prediction. The inner and outer closed loops are coupled through the collaborative optimization module to avoid correction conflicts that may be caused by single feedback, ensuring that the system maintains optimal performance throughout its entire life cycle. Attached Figure Description

[0015] Figure 1 This is a step diagram of the multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer, according to an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] 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.

[0018] 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.

[0019] 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.

[0020] Please see Figure 1 The diagram shown illustrates the steps of a multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer, according to an embodiment of the present invention.

[0021] This invention provides a multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer, comprising: Step S1: Obtain mission phase information of the UAV and simultaneously collect piezoelectric sensing signals and fiber optic grating sensing signals at key locations of the optoelectronic platform structure. Step S2: The piezoelectric sensing signal and the fiber optic grating sensing signal are fused in the time and frequency domains, and a composite feature description is generated by combining the task stage information. Step S3, performing collaborative scheduling based on the composite feature description, includes: classifying the current disturbance into a pattern and selecting the corresponding compensation algorithm based on the classification result; predicting the disturbance trend in the future time domain based on the task stage information and scheduling computing resources in an event-triggered manner and activating the compensation algorithm. Step S4: Obtain the compensated optical axis residual error and the deviation between the actual disturbance and the predicted disturbance; Step S5: Correct the parameters of the time-frequency domain feature fusion and the discrimination boundary of the pattern classification based on the optical axis residual error; Step S6: Based on the deviation between the actual disturbance and the predicted disturbance, correct the feature library parameters of the task stage information and the trigger threshold of the event-triggered scheduling.

[0022] Specifically, piezoelectric sensing signals and fiber optic grating sensing signals at key locations of the optoelectronic platform structure are collected synchronously by piezoelectric sensors and fiber optic grating sensors: (1) Piezoelectric thin film sensors are selected and are attached to three key locations: the optoelectronic platform and the body connection bracket, the gimbal motor stator housing and the camera lens base. The sensor output charge signal is converted into a 0-5V voltage signal by a charge amplifier and sent to the common signal processing circuit; (2) Fiber Bragg grating (FBG) sensors (center wavelength 1550nm, grating length 5mm) are selected and are attached to the back of the same key location with epoxy resin glue (to avoid spatial conflict). The FBG is demodulated by a fiber demodulator (sampling rate 1kHz) to demodulate the wavelength change and output a voltage signal proportional to the low-frequency deformation, which is also sent to the common signal processing circuit. In addition, the shared signal processing circuit uses a multi-channel synchronous sampling ADC (sampling rate 2kHz) to simultaneously acquire piezoelectric signals and FBG demodulated signals. After being timestamped by the FPGA, the data is packaged into data frames and transmitted to the main control processor via the SPI bus. In implementation, the main control processor communicates with the flight control computer via the CAN bus to receive the current mission phase identifier of the UAV in real time. The flight control system divides the flight phase into: hovering phase, forward flight phase, turning phase, and landing phase. Each phase is updated in the form of enumerated values ​​with an update cycle of 20ms. The electro-optical pod has a built-in three-axis fiber optic gyroscope to measure the line-of-sight angular velocity, integrates it to obtain the line-of-sight angular displacement, and uses camera image processing algorithms to extract the offset of image feature points. The two are fused to obtain the residual error angle of the optical axis relative to inertial space (optical axis residual error), with a sampling rate of 1kHz.

[0023] Specifically, the process of generating composite feature descriptions: (1) Generation of disturbance trend prediction signal: The main control processor has a pre-set disturbance spectrum feature library, which is established through multiple flight experiments. That is, the vibration acceleration spectrum of each measuring point of the photoelectric platform is collected in each mission phase, and the dominant frequency band (frequency band with energy ratio > 15%) and its average energy level are statistically obtained. In implementation, the current mission phase identifier is input, and the dominant frequency band and corresponding energy level in the future time domain (i.e., the next 1-3 seconds) are output (this value is dimensionless and normalized to 0-1); a combination of table lookup and linear extrapolation is used. If the mission phase is about to switch (as predicted by the flight control), the characteristics of the next phase are preloaded in advance. (2) Real-time sensing feature vector extraction: High-frequency feature extraction: After the piezoelectric signal passes through a bandpass filter (100-500Hz), the root mean square value is calculated, and wavelet packet decomposition is used to extract the energy of each frequency band in the fourth layer as a high-frequency feature vector (8-dimensional). Low-frequency feature extraction: After the fiber optic grating signal passes through a low-pass filter (0-50Hz), the time-domain mean, variance, peak value and main frequency component are extracted as low-frequency feature vectors (6-dimensional). Weighted fusion process: Real-time perceived feature vector = [α·high frequency feature vector, (1-α)·low frequency feature vector], where α is the fusion weight coefficient, the initial value α = 0.5, and the value range is [0.3, 0.7]. (3) Spatiotemporal alignment and composite feature description: The perturbation trend prediction signal (this value is a scalar representing the predicted dominant frequency band category) is concatenated with the real-time sensing feature vector (14-dimensional) and aligned in time dimension. That is, the prediction signal is updated every 20ms, and the sensing feature is calculated every 2ms. The latest prediction signal is reused in each control cycle (2ms) to form a 15-dimensional composite feature description vector.

[0024] Specifically, the composite feature description is input into a pre-stored one-dimensional convolutional neural network to classify the current disturbance into a pattern and output the classification result. The input parameter of the one-dimensional convolutional neural network is the composite feature description vector (15-dimensional), and the output parameter is a probability distribution of four types (low frequency dominant, mid frequency dominant, high frequency dominant, and multi-frequency mixed). The network architecture of the convolutional neural network in this embodiment is as follows: Input layer → Convolutional layer 1 (8 1×3 convolutional kernels, stride 1, ReLU) → Pooling layer 1 (1×2 max pooling) → Convolutional layer 2 (16 1×3 convolutional kernels, stride 1, ReLU) → Global average pooling → Fully connected layer (32 nodes, ReLU) → Output layer (4 nodes, Softmax); In addition, it is trained using offline collected labeled data (manually labeled perturbation types) until the classification accuracy is >95%; In implementation, if the energy proportion of the 0.5-5Hz frequency band exceeds the first threshold, a low-frequency dominant mode is output; if the energy proportion of the 5-50Hz frequency band exceeds the second threshold, a mid-frequency dominant mode is output; if the energy proportion of the 50-500Hz frequency band exceeds the third threshold, a high-frequency dominant mode is output; if the energy proportion of at least two frequency bands exceeds their respective thresholds, a multi-frequency hybrid mode is output. It is understood that the first, second, and third thresholds are usually calibrated based on engineering experience or flight experiments. Generally, the first threshold is set to 15%, the second threshold to 20%, and the third threshold to 10%.

[0025] Specifically, the corresponding compensation algorithm is selected based on the classification results, including: If the classification result is a low-frequency dominant mode, the sliding mode disturbance observer is selected as the compensation algorithm. The control law includes an equivalent control and switching term to suppress large disturbances of 0.5-5Hz. If the classification result is a mid-frequency dominant mode, the improved repetitive control disturbance observer is selected as the compensation algorithm, which contains a periodic signal generator and is designed for periodic disturbances of 10-40Hz. If the classification result is a high-frequency dominant mode, the piezoelectric active actuator feedforward compensation algorithm is selected as the compensation algorithm, and the piezoelectric stack (PZT) is driven to generate reverse displacement in real time according to the high-frequency component. If the classification result is a multi-frequency mixed mode, a weighted parallel disturbance observer is selected as the compensation algorithm. The weighted parallel disturbance observer is composed of a sliding mode disturbance observer, an improved repetitive control disturbance observer, and a piezoelectric active actuator feedforward compensation algorithm connected in parallel according to preset weights. In implementation, the preset weights of the sliding mode disturbance observer, the improved repetitive control disturbance observer, and the piezoelectric active actuator are 0.4, 0.3, and 0.3, respectively. It is understandable that selecting different compensation algorithms based on the disturbance mode classification results is based on the essential differences in the physical characteristics of disturbances and control requirements in each frequency band: Low-frequency disturbances (0.5-5Hz) mainly manifest as large-amplitude, non-periodic disturbances such as rigid body motion and ship swaying. Their energy is concentrated, changes slowly but has large amplitude. Sliding mode disturbance observers force the system state to move along the sliding surface through discontinuous control terms, exhibiting complete robustness to parameter perturbations and external disturbances, especially suitable for suppressing low-frequency large-amplitude disturbances, and are independent of disturbance periodicity. Mid-frequency disturbances (5-50Hz) mainly originate from the rotor / propeller rotation fundamental frequency and its harmonics, exhibiting strict periodicity. Improved repetitive control disturbance observers utilize the internal model principle to embed the periodic signal model into the controller, enabling them to effectively address already... Knowing the fundamental frequency and its harmonic components, zero steady-state error suppression is achieved, with a suppression depth far exceeding that of aperiodic controllers. High-frequency disturbances (50-500Hz) mainly manifest as structural resonance and piezoelectric hysteresis nonlinearity, with wavelengths comparable to structural dimensions. Rigid body control theory fails in this frequency band. Piezoelectric active actuator feedforward compensation directly generates reverse deformation at the structural level, offsetting high-frequency vibrations from the energy source. The response speed can reach the microsecond level, making it the only effective means to handle high-frequency resonance. In multi-frequency hybrid mode, a single algorithm cannot cover the entire frequency band. Therefore, a weighted parallel structure is adopted, with each algorithm compensating for its preferred frequency band. Through weight allocation, division of labor and cooperation in the frequency domain are achieved, avoiding the phase lag and stability contradictions faced by a single algorithm in the full-frequency band design.

[0026] Specifically, if the predicted dominant frequency band in the future time domain (1 second in the future) is different from the current one or the energy level exceeds a preset energy threshold (usually 0.7), a pre-activation event is generated. In response to its generation, the target compensation algorithm is switched from low-power standby (only retaining the basic DOB) to full-function state 100ms in advance, and the CPU core frequency is locked at 1.2GHz. If the predicted energy level in the future time domain (2 seconds in the future) is lower than a preset calm threshold (usually 0.3), a low-power event is generated. In response to its generation, the current compensation algorithm is switched to the basic disturbance observer (basic third-order linear DOB), and the CPU core frequency is reduced to 400MHz to release excess memory. If the Euclidean distance between the real-time calculated real-time sensing feature vector and the predicted feature vector exceeds a preset deviation threshold (usually 0.2, which is the normalized distance value), a rescheduling event is immediately generated to re-run the mode classification and allocate resources. Understandably, before the task phase switch or a sudden increase in disturbance energy, the trend of disturbance change is known in advance through task phase prediction. Since there is a startup delay (including code loading, cache warm-up, CPU frequency increase, etc., about 50-100ms) when the algorithm switches from low-power standby to full-function state, if compensation is only started after the disturbance arrives, a response blind zone will be generated. The pre-activation event triggers resource allocation in advance based on the prediction results, so that the compensation algorithm is in a ready state before the disturbance arrives, eliminating the response delay. The energy threshold Ten is set (0.7) according to the disturbance energy distribution under typical working conditions to ensure that it is only activated when high-performance compensation is really needed, avoiding frequent switching. When the predicted disturbance energy in the future time domain is lower than the calm threshold (0.3), it means that the disturbance has little impact on the optoelectronic platform, and the basic linear disturbance observer can meet the stability requirements. At this time, it will switch to Low power mode, by shutting down complex algorithm modules and reducing CPU frequency, can significantly reduce power consumption and heat generation. Based on the resource management concept of supply on demand, it actively gives up computing resources during the disturbance calm period to reserve computing power for other airborne tasks (such as image compression and target recognition). There is always a possibility of deviation between the prediction model and the actual working conditions (such as sudden wind or flight control command changes). When the Euclidean distance between the real-time sensing feature vector and the prediction signal exceeds the deviation threshold Dth, it indicates that the current prediction has been significantly inaccurate. Scheduling decisions based on inaccurate predictions may lead to a mismatch between the compensation algorithm and the actual disturbance. At this time, rescheduling must be triggered immediately. Based on the latest sensing information, the mode classification and resource allocation are re-performed to ensure the robustness of the system with a safety redundancy mechanism. The setting of the deviation threshold Dth (0.2) is determined based on the statistical distribution of prediction errors in a large number of flight experiments, taking into account both trigger sensitivity and noise resistance. Specifically, the nominal model of the third-order linear DOB is derived from the rigid body dynamics of the optoelectronic platform. It runs continuously in low-power mode with a CPU utilization of about 5%, which is much lower than that of sliding mode DOB (CPU utilization of about 18%), RC-DOB (CPU utilization of about 15%) and weighted parallel DOB (CPU utilization of about 35%).

[0027] Specifically, based on the parameters fused from the time-frequency domain features corrected for optical axis residual error and the discrimination boundary for mode classification: the power spectral density (PSD) of the optical axis residual error is calculated in real time, updated every 0.5 seconds using the Welch method, and the energies in the 0.5-5Hz, 5-50Hz, and 50-500Hz frequency bands are integrated and recorded as low-frequency residual energy values, mid-frequency residual energy values, and high-frequency residual energy values, including: If the residual energy value in the low-frequency band is greater than the low-frequency threshold (low-frequency threshold = 0.05 mrad² / Hz), then increase the fusion weighting coefficient α: α←α+Δα, Δα = 0.05, but the upper limit of α is 0.7; at the same time, the cutoff frequency f of the low-frequency filter is adjusted. low Shift to lower: f low ←f low -2Hz, but its lower limit is 5Hz; If the residual energy value in the mid-frequency band is greater than the mid-frequency threshold (mid-frequency threshold = 0.03 mrad² / Hz), then adjust the classification threshold of the mid-frequency dominant mode in the one-dimensional convolutional neural network to reduce the probability threshold of this class in the Softmax output from 0.5 to 0.45, making it easier to be selected; If the residual energy value in the high-frequency band is greater than the high-frequency threshold (high-frequency threshold = 0.02 mrad² / Hz), then increase the fusion weighting coefficient α and set the high-frequency filter cutoff frequency f. high Upward offset: f high ←f high +5Hz, but its upper limit is 500Hz.

[0028] Specifically, the feature library parameters of the task stage information and the trigger threshold of the event-triggered scheduling are corrected based on the deviation between the actual disturbance and the predicted disturbance: (1) Calculate the deviation vector ΔS between the actual disturbance spectrum and the dominant frequency band in the prediction feature library for that stage every 0.5 seconds (normalized frequency band energy difference); (2) Update the feature library: the energy level of the dominant frequency band = 0.8 × old measured value + 0.2 × new measured value; if the deviation continues to exceed 10%, the feature library will be relearned. (3) Monitor the deviation Δt between the actual triggering time and the ideal triggering time of the pre-activation event. If |Δt|>50ms, adjust the energy threshold Ten proportionally: Ten←Ten+0.02·sign(Δt), with a limit of 0.5-0.9. (4) Monitor the rescheduling event trigger frequency f trigger If f trigger If the frequency is greater than 0.2Hz (i.e., triggered more than once every 5 seconds), the deviation threshold Dth will be increased: Dth←Dth+0.01, with an upper limit of 0.3.

[0029] Specifically, during collaborative optimization: (1) Input: The set of correction values ​​(Δα, Δf) of the first feedback branch in the past hour. low ,Δf high The set of correction values ​​(ΔTen, ΔDth, ΔS) of the second feedback branch. (2) Optimization objective: Minimize the root mean square error (RMSE) of the optical axis residual error, and at the same time minimize the computational resource utilization (CPU utilization); (3) Optimization algorithm: Multi-objective Bayesian optimization, using Gaussian process as surrogate model, and expected improvement (EI) as acquisition function; (4) Decision variables: α, f low f high The system comprises 7 dimensions: Ten, Dth, and the classification thresholds for each category in a one-dimensional convolutional neural network. (5) Optimize once per flight cycle (about 1 hour), update the Gaussian process after collecting new data points, output the Pareto front, and select the compromise solution as the initial parameters for the next cycle; (6) Output: The global optimal matching parameter set is sent to the fusion sensing unit, the collaborative scheduling unit and the dual feedback correction unit to realize cross-layer collaboration.

[0030] 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.

[0031] 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 multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer, characterized in that, include: The mission phase information of the UAV is obtained, and piezoelectric sensing signals and fiber optic grating sensing signals at key locations of the optoelectronic platform structure are collected simultaneously. The time-frequency domain features of the piezoelectric sensing signals and fiber optic grating sensing signals are fused, and a composite feature description is generated by combining the mission phase information. The collaborative scheduling based on the composite feature description includes: classifying the current disturbance into a pattern and selecting the corresponding compensation algorithm based on the classification result; predicting the future disturbance trend in the time domain based on the task stage information and scheduling computing resources in an event-triggered manner and activating the compensation algorithm. Obtain the compensated optical axis residual error, obtain the deviation between the actual disturbance and the predicted disturbance, correct the parameters of the time-frequency domain feature fusion and the discrimination boundary of the pattern classification based on the optical axis residual error, and correct the feature library parameters of the task stage information and the trigger threshold of the event-triggered scheduling based on the deviation between the actual disturbance and the predicted disturbance.

2. The multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer according to claim 1, characterized in that, The process of generating composite feature descriptions includes: The flight control system obtains the mission phases of the UAV in real time, including the hovering phase, forward flight phase, turning phase, and carrier landing glide phase. Based on the task stage, the dominant frequency bands and energy levels of the corresponding stage are extracted from the preset disturbance spectrum feature library to generate a disturbance trend prediction signal; The high-frequency stress wave signal sensed by the piezoelectric sensor and the low-frequency deformation signal sensed by the fiber optic grating sensor are synchronously acquired through a shared signal processing circuit. High-frequency features and low-frequency features are extracted separately, and the high-frequency features and low-frequency features are weighted and fused to obtain a real-time sensing feature vector. Align the disturbance trend prediction signal with the real-time sensing feature vector in the time dimension, and determine the aligned composite data as the composite feature description.

3. The multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer according to claim 1, characterized in that, The parameters for time-frequency domain feature fusion include the fusion weighting coefficients of the piezoelectric sensing signal and the fiber optic grating sensing signal, the filter cutoff frequency for high-frequency feature extraction, and the filter cutoff frequency for low-frequency feature extraction.

4. The multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer according to claim 2, characterized in that, The composite feature description is input into a pre-stored one-dimensional convolutional neural network to perform pattern classification on the current perturbation and output the classification result; The classification results include low-frequency dominant mode, mid-frequency dominant mode, high-frequency dominant mode, and multi-frequency mixed mode. Specifically, the low-frequency dominant mode corresponds to an energy proportion in the 0.5-5Hz frequency band exceeding a first threshold, the mid-frequency dominant mode corresponds to an energy proportion in the 5-50Hz frequency band exceeding a second threshold, the high-frequency dominant mode corresponds to an energy proportion in the 50-500Hz frequency band exceeding a third threshold, and the multi-frequency hybrid mode corresponds to an energy proportion in at least two frequency bands exceeding their respective thresholds.

5. The multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer according to claim 4, characterized in that, Based on the classification results, a corresponding compensation algorithm is selected, including: If the classification result is a low-frequency dominant mode, the sliding mode perturbation observer is selected as the compensation algorithm. If the classification result is a mid-frequency dominant mode, the improved repetitive control disturbance observer is selected as the compensation algorithm; If the classification result is a high-frequency dominant mode, the piezoelectric active actuator feedforward compensation algorithm is selected as the compensation algorithm. If the classification result is a multi-frequency mixed mode, a weighted parallel disturbance observer is selected as the compensation algorithm. The weighted parallel disturbance observer is composed of a sliding mode disturbance observer, an improved repetitive control disturbance observer, and a piezoelectric active actuator feedforward compensation algorithm connected in parallel according to preset weights.

6. The multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer according to claim 5, characterized in that, Based on the task stage information, predict future time-domain disturbance trends to determine event triggering methods, and schedule computing resources and activate the compensation algorithm according to the event triggering methods, including: If it is predicted that a sudden change in the frequency band of the disturbance or the energy level of the disturbance will exceed the preset energy threshold in the future time domain, a pre-activation event is generated, and in response to the generation of the pre-activation event, the corresponding compensation algorithm is switched from the low-power standby state to the running state before the disturbance arrives, and the corresponding computing resources are allocated. If the predicted disturbance energy level in the future time domain is lower than a preset calm threshold, a low-power event is generated, and in response to the low-power event, the current compensation algorithm is switched to the basic disturbance observer, and excess computing resources are released. If the deviation between the real-time perceived feature vector and the disturbance trend prediction signal exceeds a preset deviation threshold, a rescheduling event is immediately generated, and in response to the rescheduling event, the compensation algorithm is re-executed to select and allocate computational resources.

7. The multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer according to claim 6, characterized in that, The basic disturbance observer is a third-order linear disturbance observer. The third-order linear disturbance observer runs continuously in low-power mode, and its computational resource utilization is lower than that of any of the sliding mode disturbance observer, the improved repetitive control disturbance observer, and the weighted parallel disturbance observer.

8. The multi-band vibration compensation control method for an UAV airborne optoelectronic platform based on a disturbance observer according to claim 7, characterized in that, The process of correcting the parameters of the time-frequency domain feature fusion and the discrimination boundary of the pattern classification based on the optical axis residual error includes: The power spectral density of the optical axis residual error after compensation is calculated in real time, and the residual energy values ​​of the low-frequency band, mid-frequency band and high-frequency band are determined respectively. The parameters of the time-frequency domain feature fusion are corrected based on the residual energy values ​​of the low-frequency, mid-frequency, and high-frequency bands, wherein: If the residual energy value in the low-frequency band is greater than the low-frequency threshold, then increase the fusion weight coefficient of the fiber grating sensing signal in the time-frequency domain feature fusion, and adjust the filter cutoff frequency during low-frequency feature extraction to shift towards the low-frequency direction. If the residual energy value in the mid-frequency band is greater than the mid-frequency threshold, then the discrimination boundary of the corresponding mid-frequency dominant mode in the one-dimensional convolutional neural network is adjusted to shift towards the direction of decreasing mid-frequency energy. If the residual energy value in the high-frequency band is greater than the high-frequency threshold, then increase the fusion weight coefficient of the piezoelectric sensing signal in the time-frequency domain feature fusion, and adjust the filter cutoff frequency during high-frequency feature extraction to shift towards the high-frequency direction.

9. The multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer according to claim 8, characterized in that, The process of correcting the feature library parameters of the task stage information and the trigger threshold of the event-triggered scheduling based on the deviation between the actual disturbance and the predicted disturbance includes: Obtain the spectral distribution of the actual disturbance and calculate the deviation vector between the actual disturbance and the disturbance trend prediction signal in the frequency domain; The deviation vector is used as a correction value to update the statistical parameters of the dominant frequency band and energy level for the corresponding task stage in the disturbance spectrum feature library. The deviation between the actual triggering time and the ideal triggering time of the pre-activation event is monitored in real time. If the deviation exceeds the time deviation threshold, the energy threshold is adjusted to correct the prediction sensitivity. The triggering frequency of rescheduling events is monitored in real time. If the triggering frequency exceeds the frequency threshold, the preset deviation threshold is adjusted to correct the error in the direction of increasing tolerance.

10. The multi-band vibration compensation control method for an unmanned aerial vehicle (UAV) airborne optoelectronic platform based on a disturbance observer according to claim 9, characterized in that, It also includes collaborative optimization steps: The correction amount based on the optical axis residual error and the correction amount based on the deviation between the actual disturbance and the predicted disturbance are input into the collaborative optimization module; The collaborative optimization module adopts multi-objective Bayesian optimization, with the optimization objectives being to minimize the optical axis residual error and the computational resource utilization rate. At the same time, it determines the global optimal match of time-frequency domain feature fusion parameters, pattern classification and discrimination boundaries, feature library parameters, and event triggering thresholds. The globally optimal match is used as the initial parameter for the next control cycle.

Images