Dynamic resonance suppression image stabilization feedback control method and system

Abstract

The invention provides a dynamic resonance suppression image stabilization feedback control method and system, and relates to the technical field of spacecraft engineering, and the method comprises the steps: driving a piezoelectric ceramic actuator or an electromagnetic actuator array according to a reverse control instruction, generating a reverse acting force which is opposite to a vibration phase and is matched with a vibration amplitude, and enabling the piezoelectric ceramic actuator or the electromagnetic actuator array to rotate; applying the reverse acting force to the imaging optical assembly to obtain the change of the vibration state of the platform; based on platform vibration state changes, platform response signals after vibration offset are collected in real time through the sensing array; and carrying out online setting on parameters of the self-adaptive controller through a closed-loop feedback mechanism by utilizing the platform response signal after vibration offset so as to realize continuous optimization of a vibration suppression effect. Through sensing acquisition, intelligent processing and self-adaptive control, a reverse acting force is generated in real time to suppress vibration of the platform, and the vibration suppression effect is continuously optimized through closed-loop feedback.

Description

Technical Field

[0001] This invention relates to the field of spacecraft engineering technology, and in particular to a dynamic resonance suppression image stabilization feedback control method and system. Background Technology

[0002] The increasing resolution of optical remote sensing satellite imaging is constantly pushing the limits of platform vibration suppression technology. Traditional vibration control methods face bottlenecks in data processing; their limitations in noise reduction accuracy, feature extraction capabilities, and adaptive algorithms make it difficult to meet the extreme requirements of next-generation high-resolution payloads for image stabilization accuracy. To suppress vibration, active vibration control (AVC) technology is widely used. This technology collects vibration signals through inertial sensors, processes them through control algorithms, and then drives actuators to generate compensating forces. However, existing technologies have significant shortcomings in data processing, severely limiting their performance. First, the original vibration signal of the platform has a low signal-to-noise ratio and is mixed with various noises. Although traditional filtering methods (such as low-pass filtering or conventional Kalman filtering) can partially suppress noise, some introduce phase lag and rely on accurate assumptions about the statistical characteristics of system noise. In the complex on-orbit time-varying environment, model mismatch may lead to a decrease in filtering accuracy or even divergence, causing the controller to generate commands based on inaccurate signals. This not only has a limited suppression effect but may also introduce secondary interference. Second, existing systems mostly adopt shallow data fusion strategies (such as complementary filtering), which only realize basic attitude calculation and fail to deeply explore the deep geometric motion characteristics (such as motion trajectory curvature, higher-order motion derivatives, etc.) contained in the vibration data. This may lead to insufficient perception of complex vibration modes and spatiotemporal correlation characteristics of the system. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a dynamic resonance suppression image stabilization feedback control method and system, which generates a reverse force in real time to suppress platform vibration through sensing acquisition, intelligent processing and adaptive control, and continuously optimizes the vibration suppression effect through closed-loop feedback.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A first aspect is a dynamic resonance suppression and image stabilization feedback control method, the method comprising: The vibration acceleration and angular velocity signals of the platform in three degrees of freedom are collected in real time by a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. The Kalman filter algorithm is used to denoise the vibration acceleration and angular velocity signals to obtain the denoised platform state signal. The denoised platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. The vibration state vector is input into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount. The corrected vibration state vector is input into an adaptive controller based on recursive least squares, and inverse control commands to counteract vibration are generated in real time through a combination of feedforward and feedback. According to the reverse control command, the piezoelectric ceramic actuator or electromagnetic actuator array is driven to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and the reverse force is applied to the imaging optical component to obtain the change in the platform vibration state. Based on the changes in the platform's vibration state, the platform's response signal after vibration cancellation is collected in real time through the sensor array. By utilizing the platform response signal after vibration cancellation, the parameters of the adaptive controller are tuned online through a closed-loop feedback mechanism to achieve continuous optimization of the vibration suppression effect.

[0005] Secondly, a dynamic resonance suppression and image stabilization feedback control system includes: The acquisition module is used to acquire the vibration acceleration and angular velocity signals of the platform in three degrees of freedom in real time through a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. The fusion module is used to perform noise reduction on the vibration acceleration and angular velocity signals using the Kalman filter algorithm to obtain the noise-reduced platform state signal; the noise-reduced platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. The correction module is used to input the vibration state vector into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount. The control module is used to input the corrected vibration state vector into the adaptive controller based on the recursive least squares method, and generate inverse control commands to counteract the vibration in real time through a combination of feedforward and feedback. The execution module is used to drive the piezoelectric ceramic actuator or electromagnetic actuator array according to the reverse control command to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and apply the reverse force to the imaging optical component to obtain the change in the vibration state of the platform. The post-cancellation acquisition module is used to acquire the platform response signal after vibration cancellation in real time through the sensor array based on the changes in the platform's vibration state. The closed-loop optimization module is used to utilize the platform response signal after vibration cancellation to tune the parameters of the adaptive controller online through a closed-loop feedback mechanism, so as to achieve continuous optimization of the vibration suppression effect.

[0006] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0007] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0008] The above-described solution of the present invention has at least the following beneficial effects: Multi-axis accelerometers and gyroscopes synchronously acquire three-degree-of-freedom vibration signals. Combined with Kalman filtering (establishing state and observation equations that fit the platform's dynamics, followed by prediction, updating, correction, and noise reduction) and quaternion-based complementary filtering fusion, accurate real-time platform vibration state vectors can be obtained, providing high-quality foundational data for subsequent processing. A deep neural network model first standardizes and preprocesses the vibration state vectors to adapt to the input, then generates a correction vector through multi-layer nonlinear transformations. This can specifically compensate for state deviations, conforming to the platform's nonlinear vibration characteristics and improving the accuracy of the state vectors. The adaptive controller initializes parameters and covariance matrix to adapt to the initial state. The feedforward channel derives components based on the platform vibration model, while the feedback channel updates parameters and calculates components in real time using recursive least squares. During fusion, weighted summation and amplitude limiting are applied to generate... The system adapts to the reverse control commands of the actuators, balancing accuracy and safety. It generates actuator drive signals based on command parameters, driving the actuators to produce a reverse mechanical force, which is applied to the optical component mounting base or vibration isolation mechanism. Through mechanical coupling, it efficiently cancels vibration and accurately acquires changes in platform vibration state. Based on these vibration state changes, it generates synchronous trigger commands, controlling sensors to synchronously acquire response signals. These signals are preprocessed to obtain a unified format digital signal, and geometric features are analyzed, calculated, and encapsulated for alignment. This yields response signals containing deep motion information and adapted to feedback, providing effective data for parameter tuning. The system extracts the characteristics and parameters of the response signals as evaluation indicators. After calculating the deviation, it uses a recursive least squares method to adjust and update the controller parameters online, driving the next cycle of operation. This allows for continuous adaptation to changes in platform vibration characteristics and optimization of the suppression effect. Attached Figure Description

[0009] Figure 1 This is a schematic flowchart of a dynamic resonance suppression image stabilization feedback control method provided by an embodiment of the present invention.

[0010] Figure 2This is a schematic diagram of a dynamic resonance suppression image stabilization feedback control system provided by an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, an embodiment of the present invention proposes a dynamic resonance suppression and image stabilization feedback control method, the method comprising the following steps: Step 1: The vibration acceleration and angular velocity signals of the platform in three degrees of freedom are collected in real time by a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. Step 2: The Kalman filter algorithm is used to denoise the vibration acceleration and angular velocity signals to obtain the denoised platform state signal; the denoised platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. Step 3: Input the vibration state vector into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector; wherein, the deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount; Step 4: Input the corrected vibration state vector into the adaptive controller based on the recursive least squares method, and generate inverse control commands to counteract the vibration in real time through a combination of feedforward and feedback. Step 5: According to the reverse control command, drive the piezoelectric ceramic actuator or electromagnetic actuator array to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and apply the reverse force to the imaging optical component to obtain the change in the platform vibration state. Step 6: Based on the changes in platform vibration state, the platform response signal after vibration cancellation is collected in real time through the sensor array; Step 7: Using the platform response signal after vibration cancellation, the parameters of the adaptive controller are tuned online through a closed-loop feedback mechanism to achieve continuous optimization of the vibration suppression effect.

[0013] In this embodiment of the invention, data is collected by a multi-axis accelerometer and gyroscope array, which can simultaneously acquire the platform's three-degree-of-freedom vibration acceleration and angular velocity signals, comprehensively covering translational and rotational vibration states. This provides complete and continuous raw data for subsequent processing, ensuring the integrity of the basic data. Kalman filtering can reduce random noise in the raw signal and improve the signal-to-noise ratio. Sensor fusion can integrate complementary features of two types of sensors, eliminate data redundancy and conflict, and ultimately form a unified real-time vibration state vector, making the vibration state description more consistent and accurate. A pre-trained deep neural network establishes a nonlinear mapping between vibration state and correction amount based on historical data, adapting to complex vibration characteristics and directly outputting the corrected vibration state vector, making the correction more in line with actual laws and providing a realistic basis for the generation of control commands. The recursive least squares method can identify the time-varying characteristics of vibration in real time. The system dynamically adapts controller parameters to operating conditions; a combination of feedforward and feedback generates inverse control commands, which can handle predictable disturbances and correct residual deviations, making the commands more targeted and comprehensive; the actuator array is driven to generate a reverse force with opposite phase and matching amplitude, which is applied to the imaging optical components, directly affecting the key vibration carrier, effectively changing the platform's vibration state, achieving active vibration cancellation, and driving the platform towards stability; the response signal after vibration cancellation is collected by the sensor array, timely acquisition of suppression effect data is obtained, providing accurate feedback for controller parameter adjustment, and ensuring the correlation between feedback data and control effect; relying on closed-loop feedback to tune controller parameters, the parameters can be dynamically adjusted according to the actual suppression effect, allowing the controller to adapt to vibration changes, continuously optimize the suppression effect, improve the system's adaptability to different operating conditions, and ensure long-term stable suppression performance.

[0014] In a preferred embodiment of the present invention, step 1 above, which involves acquiring the vibration acceleration and angular velocity signals of the platform in three degrees of freedom in real time through a sensor array composed of multi-axis accelerometers and gyroscopes deployed on the imaging platform, includes: Step 11: Deploy a three-axis MEMS accelerometer and a fiber optic gyroscope on the rigid structure of the imaging platform to form a multi-sensor array. Perform coordinate system calibration and initial calibration for each sensor to obtain the calibration results. Specifically, prioritize high-reliability models that meet the selection principles for low-cost satellite EEE components. The three-axis MEMS accelerometer must meet a measurement accuracy of 0.1 mg, and the fiber optic gyroscope must meet a zero-bias stability of no more than 0.01 degrees per hour. Prioritize mature products with flight experience or automotive-grade or higher. Deploy the two types of sensors in key vibration-sensitive areas of the 0.25m rigid structure of the camera imaging platform, such as near the primary mirror assembly and focal plane assembly. At the support structure, a rigid structure is used to reduce deformation interference during vibration signal transmission. Using the camera coordinate system (Cartesian right-hand coordinate system, with the +Z direction as the optical axis) as the reference, a cubic prism is used as the calibration reference. The installation angle deviation of each sensor is calibrated by a laser interferometer, and the transformation relationship between the sensor local coordinate system and the platform global coordinate system is established. Initial calibration of the sensors is carried out. A high-precision vibration table is used to apply a standard acceleration signal with a known amplitude to the MEMS accelerometer, the output deviation is recorded and the error compensation coefficient is calculated. A turntable is used to perform multi-attitude zero-bias tests on the fiber optic gyroscope to determine the zero-bias error correction value. Finally, a calibration result including the coordinate system transformation matrix and sensor error compensation coefficient is formed.

[0015] Step 12: Based on the sensor calibration results, the accelerometer is used to collect the linear vibration acceleration signals of the platform in the X, Y, and Z axes in real time, and the gyroscope is used to collect the angular velocity signals of the platform around the X, Y, and Z axes in real time. Specifically, this includes: extracting the coordinate system transformation matrix from the sensor calibration results and preloading it into the data acquisition control unit of the camera focal plane circuit; starting the MEMS accelerometer and collecting the linear vibration acceleration signals of the platform in the X, Y, and Z axes in real time based on the calibration parameters. During the acquisition process, the accelerometer's operating temperature is stabilized at -20°C using the camera's thermal control components. Within the cabin electronic equipment temperature range of up to 55℃, temperature drift affecting measurement accuracy is avoided; fiber optic gyroscopes are simultaneously activated to collect angular velocity signals of the platform around the X, Y, and Z axes in real time based on the same calibration parameters; the temperature fluctuation of the gyroscope's working environment is maintained within ±0.5℃ by the thermal control components to ensure its zero-bias stability is maintained at 0.01 degrees per hour; the sampling frequency of the two types of sensors is set to no less than 500Hz to cover the platform's vibration bandwidth of 0.1 to 200Hz, satisfying the Nyquist sampling theorem to ensure complete capture of the dynamic changes in linear acceleration and angular velocity.

[0016] Step 13 involves synchronously sampling and formatting the vibration acceleration and angular velocity signals to eliminate timing deviations and dimensional differences between sensors, resulting in time-aligned vibration acceleration and angular velocity signals. Specifically, this includes: introducing a GNSS second pulse signal conforming to the camera communication interface requirements as a synchronization trigger source. This signal meets the RS422 level specification. Using the pulse's leading edge at the integer second as a reference, the sampling timing of the MEMS accelerometer and fiber optic gyroscope is synchronously controlled to ensure that the time deviation between single samplings of the two types of sensors does not exceed 20 μs; and building a signal format conversion module in the data processing unit of the focal plane circuit to convert the m / s output from the MEMS accelerometer... 2 The linear acceleration signal and the angular velocity signal in the rad / s range output by the fiber optic gyroscope are uniformly converted into 16-bit digital signals based on the International System of Units (SI) to eliminate dimensional differences. The converted digital signals are then subjected to time-series alignment processing. By comparing the timestamps of each frame of the signal, misaligned data is eliminated. Extreme outliers are identified and removed using the sliding window method. The final output is a vibration acceleration and angular velocity signal with consistent timestamps, uniform dimensions, and no obvious anomalies.

[0017] In this embodiment of the invention, a sensor array consisting of a three-axis MEMS accelerometer and a fiber optic gyroscope is deployed on the rigid body of the imaging platform to complete coordinate system calibration and initial calibration. This establishes a unified coordinate reference, eliminates initial errors such as sensor installation deviations, and ensures the reliability of the data acquired by the sensor array. Based on the calibration results, linear vibration acceleration signals along the X, Y, and Z axes and angular velocity signals around the three axes are acquired, comprehensively obtaining the three-degree-of-freedom translational and rotational vibration data of the platform, ensuring complete vibration data coverage. Synchronous sampling and format unification of the signals eliminate timing deviations and dimensional differences between sensors, obtaining time-aligned vibration signals, ensuring data timing consistency and the feasibility of subsequent integration and processing.

[0018] In a preferred embodiment of the present invention, step 2 above involves using a Kalman filter algorithm to denoise the vibration acceleration and angular velocity signals to obtain a denoised platform state signal; the denoised platform state signal is then fused using a sensor fusion algorithm to obtain a real-time vibration state vector of the platform, including: Step 21: Preprocess the vibration acceleration and angular velocity signals to obtain preprocessed signals; based on the preprocessed signals, establish a Kalman filter state equation describing the dynamic characteristics; according to the state equation and the sensor measurement principle, construct the Kalman filter observation equation, specifically including: firstly, preprocessing the time-aligned vibration acceleration and angular velocity signals, using the sliding window method to remove pulse-type extreme outliers, and simultaneously adjusting the signal amplitude to a range suitable for subsequent filtering according to the signal processing requirements of the camera focal plane circuit, to obtain the preprocessed signal; then, based on the preprocessed signal and combined with the dynamic characteristics of the rigid structure of the 0.25m camera imaging platform, establish the Kalman filter state equation; this state equation describes the dynamic evolution process of the system, and its state vector includes parameters such as the position, velocity, and linear acceleration of the platform's three-degree-of-freedom translational motion, and the attitude angle and angular velocity of the rotational motion; the specific form of the state equation is constructed based on the Newton-Euler equation, and by incorporating the platform's inertial parameters and kinematic relationships, the system state changes over time in the form of a state transition matrix. During calculation, the state estimate of the previous moment and the current control input are used to predict the system state at the current moment through the state transition matrix.

[0019] Simultaneously, a Kalman filter observation equation is constructed based on the sensor measurement principle. The observation equation establishes the mathematical relationship between the system state and the sensor measurement values, and its specific form takes into account the actual measurement characteristics of the selected triaxial MEMS accelerometer and fiber optic gyroscope. The observation equation incorporates the sensor calibration results, such as the accelerometer's scale factor error and the gyroscope's zero bias error compensation coefficient, to ensure that the observation model can accurately reflect the mapping between the sensor output and the real physical quantity. During calculation, the predicted system state quantity is mapped to the sensor measurement space through the observation matrix to obtain the predicted observation value. Finally, by combining the state equation and the observation equation, a complete Kalman filter framework is formed.

[0020] Step 22: Based on the state equation and observation equation, through the state prediction stage, the platform state vector and covariance matrix at the current moment are predicted by using the state estimate and covariance matrix of the previous moment, combined with the platform dynamic characteristics described by the state equation. Specifically, this includes: based on the state equation and observation equation established in step 21, entering the state prediction stage; retrieving the platform state estimate (including linear acceleration, angular velocity, etc.) and its covariance matrix of the previous moment from the camera data storage unit, the initial value of the covariance matrix is ​​determined based on the sensor's factory noise parameters and calibration error; combining the platform dynamic characteristics described by the state equation, such as the inertial delay when rigid structures transmit vibration, and the dynamic parameter drift suppressed by the thermal control components (maintaining the equipment temperature from -20℃ to 55℃), the platform state vector at the current moment is calculated recursively; simultaneously, the covariance matrix is ​​updated according to the noise characteristics of the state equation to reflect the uncertainty in the prediction process and ensure that the prediction result conforms to the actual motion law of the platform.

[0021] Step 23: Based on the platform state vector and covariance matrix, calculate the Kalman gain matrix through the observation update stage; use the gain matrix and the actual observation value at the current moment to correct the predicted state and covariance matrix, obtain the final state estimate and the updated covariance matrix, and obtain the denoised platform state signal. Specifically, this includes: based on the platform state vector and covariance matrix predicted in Step 22, combined with the statistical characteristics of sensor measurement noise (determined by the calibration results and factory parameters of the MEMS accelerometer and fiber optic gyroscope), calculate the Kalman gain matrix, which can dynamically balance the reliability of the predicted state and the credibility of the observation value; obtain the actual observation value (preprocessed acceleration and angular velocity signals) at the current moment from the sensor real-time acquisition channel, and use the Kalman gain matrix to weight and fuse the actual observation value and the predicted state; correct the predicted state and covariance matrix to eliminate some noise interference, obtain the final state estimate and the updated covariance matrix; output the final state estimate according to the signal format requirements of the camera focal plane circuit to obtain the denoised platform state signal.

[0022] Step 24: For the denoised platform state signal, a quaternion-based complementary filtering algorithm is used for multi-source information fusion and attitude calculation, directly outputting the real-time vibration state vector of the platform. Specifically, this includes: using the denoised platform state signal obtained in step 23, a quaternion-based complementary filtering algorithm is used for multi-source information fusion. The quaternion definition fits the camera coordinate system (Cartesian right-hand coordinate system, with the +Z direction as the optical axis) to avoid singularity problems in attitude calculation; during the fusion process, the measurement stability of the MEMS accelerometer in the low-frequency band is used to compensate for the zero-bias drift of the fiber optic gyroscope during long-term operation, while the fast response characteristics of the fiber optic gyroscope in the high-frequency band are used to supplement the response lag of the accelerometer under dynamic vibration; attitude calculation is performed based on the fused information, and the calculation process refers to the coordinate system correlation in the camera optomechanical main body design to obtain the real-time attitude angle and vibration parameters of the platform; finally, the real-time vibration state vector of the platform containing three-degree-of-freedom translation (X, Y, Z-axis displacement, velocity, acceleration) and rotation (angular velocity and attitude angle around the X, Y, Z axes) information is directly output.

[0023] In this embodiment of the invention, preprocessing of vibration acceleration and angular velocity signals can improve the quality of the initial signal. A Kalman filter state equation is established based on the preprocessed signal and its dynamic characteristics. An observation equation is constructed by combining the sensor measurement principle, allowing the filter model to conform to the actual motion law of the platform and the sensor characteristics. In the state prediction stage, the current state vector and covariance matrix are calculated using the previous state estimate and covariance matrix, combined with the platform's dynamic characteristics. This allows for prediction based on the platform's motion law, while the covariance matrix reflects prediction uncertainty, ensuring the rationality and consistency of the prediction results. Calculating the Kalman gain matrix dynamically balances the weights of the predicted state and the actual observations. Correcting the predicted state and covariance matrix using the gain matrix and actual observations reduces the impact of noise on state estimation, resulting in a more accurate final state estimate and a platform state signal with better noise reduction. A quaternion-based complementary filtering algorithm avoids singularity problems in attitude calculation. Simultaneously, integrating the denoised multi-source platform state signals fully leverages the complementary advantages of the accelerometer and gyroscope, efficiently completing information fusion and attitude calculation, and outputting a real-time vibration state vector that comprehensively reflects the platform's motion.

[0024] In a preferred embodiment of the present invention, step 3 above involves inputting the vibration state vector into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction value, including: Step 31: Obtain a pre-trained deep neural network model. This model is used to establish a nonlinear mapping relationship from vibration state to its correction value. Specifically, this includes: obtaining the deep neural network model, whose construction requires first determining the appropriate model architecture. Since it is necessary to process the three-degree-of-freedom temporal vibration data of the 0.25m camera imaging platform, an architecture capable of capturing dynamic temporal features is selected. At the same time, combined with the hardware interface characteristics of the camera's intelligent supercomputing-driven image depth analysis engine, the model's input layer dimension is designed to be consistent with the platform's real-time vibration state vector (including X, Y, and Z-axis displacement, velocity, linear acceleration, and angular velocity and attitude angle around the three axes), and the output layer dimension is consistent with the vibration state correction vector (corresponding to each vibration parameter). The model structure is matched with the deviation compensation amount to ensure that it can efficiently process time-series vibration data and directly adapt to the supercomputing engine hardware environment without additional adjustments. Before model training, historical data covering typical camera vibration scenarios must be collected, including vibration data caused by thermal deformation of electronic equipment in the ground simulation cabin from -20°C to 55°C, resonant vibration data transmitted by the rigid support of the structure, and real three-degree-of-freedom translational and rotational vibration data collected by a three-axis MEMS accelerometer (0.1mg level accuracy) and fiber optic gyroscope (zero bias stability less than or equal to 0.01° / h). The collected historical data is preprocessed to remove pulse-type extreme outliers to ensure data validity.

[0025] During training, preprocessed historical vibration data is used as model input, and the corresponding vibration state correction (calculated based on the actual dynamic error of the platform) is used as output. Simultaneously, platform inertial parameters such as rigid structure mass and moment of inertia, as well as actuator nonlinear characteristics such as piezoelectric ceramic hysteresis, are incorporated into the training process. Through iterative optimization of the parameters of the multi-layer nonlinear transformation within the model, the model gradually learns and establishes a nonlinear mapping relationship between vibration state and correction, ensuring that this mapping conforms to the actual dynamic laws of the camera. After model training, stability verification is conducted in the hardware environment of the camera's intelligent supercomputing-driven image depth analysis engine. The smoothness of the model's operation and output accuracy when continuously processing vibration data are tested, confirming that the model error meets the requirements of the dynamic resonance suppression image stabilization system and that there are no hardware adaptation issues. Finally, a trained and verified deep neural network model is formed. This model requires no additional adaptation and can be directly called in the supercomputing engine hardware environment for subsequent calculations of the corresponding correction from the vibration state vector.

[0026] Step 32: Standardize the real-time vibration state vector of the platform to obtain a standardized vector that conforms to the input format of the deep neural network model. Specifically, this includes: extracting key parameters from the real-time vibration state vector, including X, Y, and Z-axis displacement, velocity, linear acceleration, angular velocity around the three axes, and attitude angle; converting each parameter into a 16-bit digital signal format according to the focal plane circuit signal processing specification; and using the mean and standard deviation of each dimension obtained from historical vibration data statistics to standardize parameters of different dimensions in the real-time vector (such as acceleration in m / s²). 2 The vectors (angular velocity units rad / s) are standardized to eliminate scale differences. During the process, a sliding window method is used to remove tiny outliers caused by instantaneous sensor interference, ensuring that the numerical range of the standardized vectors is compatible with the input layer of the deep neural network model and avoiding exceeding the model's calculation threshold.

[0027] Step 33: Input the standardized vector into the deep neural network model. Through multi-layer nonlinear transformation calculations within the model, obtain the corresponding vibration state correction vector. Specifically, after inputting the standardized vector into the deep neural network model, and relying on the computing power provided by the camera intelligent supercomputing engine, complete the calculation through multi-layer nonlinear transformations from the model input layer, hidden layer to output layer. Each hidden layer extracts different features of the vibration data, such as capturing high-frequency resonance signal features in the shallow layer and exploring low-frequency vibration coupling features caused by thermal deformation in the deep layer. Furthermore, the transformation parameters of each layer are trained and optimized based on historical vibration data to adapt to the coupling characteristics of the platform's three-degree-of-freedom vibration. After calculation, directly output the vibration state correction vector with the same dimension as the input vector. Each element of the correction vector corresponds to the deviation compensation amount of displacement, velocity, acceleration, and angular velocity in the real-time vibration state.

[0028] Step 34: Based on the vibration state correction vector, compensate and correct the real-time vibration state vector of the platform to obtain the corrected vibration state vector. Specifically, this includes: based on the vibration state correction vector, performing targeted compensation on each parameter in the real-time vibration state vector of the platform, such as using the acceleration compensation amount in the correction vector to offset the sensor zero bias error (derived from the calibration result in step 11), and using the angular velocity compensation amount to correct the dynamic drift caused by thermal deformation; the compensation and correction process refers to the coordinate system relationship of the camera's optical engine main body design (Cartesian right-hand coordinate system, with +Z direction as the optical axis direction) to ensure that the correction amount in each dimension is accurately matched with the real-time vector parameters; the final output corrected vibration state vector must meet the input accuracy requirements of the subsequent adaptive controller based on the recursive least squares method, and its parameter deviation must be controlled within the allowable range of the camera dynamic resonance suppression and stabilization system.

[0029] In this embodiment of the invention, a deep neural network model trained on historical vibration data is obtained. This model has a pre-established nonlinear mapping between vibration state and correction amount, which can directly adapt to the vibration characteristics of the 0.25m camera imaging platform without re-modeling and calculation, saving data processing time. At the same time, the model training data covers common vibration scenarios of the platform and can fit the nonlinear laws of actual vibration. Standardization preprocessing of the platform's real-time vibration state vector can unify the scale of data of different dimensions (such as linear acceleration and angular velocity) and eliminate dimensional differences. At the same time, the data format matches the input requirements of the deep neural network model, avoiding the impact of inconsistent data range or format on model calculation. After the standardized vector is input into the model, through multi-layer nonlinear transformation calculation within the model, the complex correlation features contained in the vibration state data can be deeply explored, such as the coupling characteristics of the platform's three-degree-of-freedom vibration. It can adapt to the nonlinear vibration changes of the camera platform caused by thermal deformation, structural vibration, etc., and output a correction vector that matches the current vibration state. Based on the vibration state correction vector, the original real-time vibration state vector is compensated and corrected, which can directly and specifically offset the deviation components in the original vector.

[0030] In a preferred embodiment of the present invention, step 4 above involves inputting the corrected vibration state vector into an adaptive controller based on the recursive least squares method, and generating inverse control commands to counteract the vibration in real time through a combination of feedforward and feedback, including: Step 41: Initialize the control parameters of the adaptive controller and the covariance matrix required for calculation using the recursive least squares method; use the corrected vibration state vector as the state input for the current control cycle. Specifically, this includes: when initializing the control parameters of the adaptive controller, referencing the inertial parameters of the rigid structure of the 0.25m camera imaging platform (e.g., main body mass approximately 290kg, moment of inertia) and actuator characteristics (piezoelectric ceramic actuator response speed less than 1ms, output density greater than 100N / kg), setting the initial gain and control cycle (matching the 8Hz integral time update frequency of the focal plane circuit); initializing the recursive least squares method... The initial value of the covariance matrix obtained by the two-square method is determined based on the noise statistical characteristics of the triaxial MEMS accelerometer (0.1mg level accuracy) and the fiber optic gyroscope (zero bias stability less than or equal to 0.01° / h) to ensure that the initial state of the matrix is ​​adapted to the error range of the vibration signal. The corrected vibration state vector (including displacement, velocity, acceleration and angular velocity around the three axes in the X, Y and Z directions) obtained in step 34 is used as the state input of the current control cycle. Before input, it is confirmed that the vector format meets the signal interface requirements of the adaptive controller (consistent with the 16-bit digital signal output format of the focal plane circuit) to avoid calculation deviation caused by format mismatch.

[0031] Step 42: Based on the state input, the feedforward control components are calculated through the feedforward control channel according to the pre-set platform vibration model. Specifically, this includes: based on the state input in step 41, calling the pre-set platform vibration model through the feedforward control channel; the construction of this model requires first determining the parameters of typical vibration scenarios of the 0.25m camera, including low-frequency vibrations caused by thermal deformation of the electronic equipment inside the cabin from -20°C to 55°C (matching the temperature range of the electronic equipment inside the cabin in the camera thermal control interface), 1 to 100Hz resonant vibrations transmitted by rigid support structures (such as the main load-bearing frame of the camera) (fitting the dynamic characteristics of the rigid structure of the optical engine body), and periodic vibrations generated by the CMG maneuver of the satellite platform (referring to the maneuverability parameters in the satellite attitude constraints); during construction, the platform inertial parameters and the actuator hysteresis compensation model (adapting to the hysteresis characteristics of the piezoelectric ceramic actuator) are combined to establish the dynamic relationship describing the generation and transmission of vibrations. At the same time, the frequency response range of the model is determined by referring to the bandwidth requirement of 0.1 to 200Hz for the dynamic resonance suppression image stabilization system to ensure that it can cover the main vibration frequencies.

[0032] During the model training phase, historical vibration data from multiple scenarios need to be collected. Ground-based temperature control equipment is used to simulate the thermal deformation environment inside the cabin, ranging from -20°C to 55°C. A high-precision vibration table is used to simulate resonant vibrations from 1 to 100 Hz. Combined with a periodic vibration device simulating CMG maneuvers, three-axis MEMS accelerometers (0.1 mg-level accuracy) and fiber optic gyroscopes (zero-bias stability less than or equal to 0.01° / h) are used to collect three-degree-of-freedom vibration data under various scenarios. Based on the collected real vibration data, the internal parameters of the model are optimized to ensure that the vibration dynamic characteristics output by the model are consistent with the actual measurement data. At the same time, the model's response capability within the 0.1 to 200 Hz bandwidth is verified to ensure the fitting accuracy of the main vibration frequencies.

[0033] The trained platform vibration model is integrated into the feedforward control channel. When invoked, it is based on the state input and performs forward derivation according to the established dynamic relationship in the model. During the derivation process, the vibration scene characteristics of the current state (such as low-frequency vibration or resonant vibration caused by thermal deformation) are substituted to calculate the feedforward control components for the known vibration. The frequency range of the components is checked simultaneously during the derivation process to ensure that the main frequency components of the current vibration are covered. The cancellation command is generated in advance to reduce the response delay and adapt to the real-time requirements of the dynamic resonance suppression and stabilization system for vibration suppression.

[0034] Step 43: Based on the state input, calculate the difference between it and the desired steady state to obtain the real-time state error; based on the real-time state error, update the control law parameters in real time using the recursive least squares method through the feedback control channel, and calculate the feedback control component. Specifically, this includes: based on the state input in step 41, using the ideal steady state (three degrees of freedom displacement and velocity approaching 0, angular velocity approaching 0) when the 0.25m camera is imaging as a benchmark, calculate the difference between the two in each dimension (X, Y, Z directions and around the three axes) to obtain the real-time state error; based on this error, call the recursive least squares method through the feedback control channel to update the control law parameters in real time, incorporating platform dynamic drift compensation during the update process, with the parameter update frequency synchronized with the control cycle (8Hz); calculate the feedback control component based on the updated control law parameters and the real-time state error, which focuses on correcting residual vibration deviations not covered by the feedforward control.

[0035] Step 44: The feedforward control component and the feedback control component are fused, and weighted summation and amplitude limiting are performed to obtain a preliminary reverse control command. Specifically, this includes: when fusing the feedforward control component and the feedback control component, the weighting coefficient is dynamically adjusted according to the current vibration scenario. When the proportion of periodic vibration is known to be high, the weight of the feedforward component is set to 0.6 to 0.8; when there are many sudden random vibrations, the weight of the feedback component is set to 0.6 to 0.8. The result after weighted summation is subjected to amplitude limiting. The limiting threshold is determined based on the maximum output of the actuator (for piezoelectric ceramic actuators, the limit is set according to the output density greater than 100 N / kg; for electromagnetic actuators, the limit is set according to the rated output force range) to avoid the command exceeding the physical capacity of the actuator and causing overload. The limiting process also refers to the rated operating parameters of the actuator (such as the maximum allowable current of the piezoelectric ceramic actuator and the rated output force range of the electromagnetic actuator) and the camera thermal control design requirements to avoid the command causing the actuator to operate beyond the rated load and cause overcurrent, or to cause overheating due to excessive heat generated by continuous high load exceeding the thermal control bearing capacity. This process yields a preliminary reverse control command.

[0036] Step 45 involves output conversion of the preliminary reverse control command to obtain the final reverse control command for driving the actuator array, which is opposite in phase to the vibration and matches the amplitude. Specifically, this includes: output conversion of the preliminary reverse control command; matching the drive signal format according to the actuator type; when driving a piezoelectric ceramic actuator, converting the command into a PWM drive signal (efficiency greater than 85%, meeting the overall efficiency requirement of over 75% for the secondary power supply); when driving an electromagnetic actuator array, converting it into a constant current signal; incorporating the actuator's phase compensation parameters during the conversion process to ensure that the force corresponding to the final command is opposite in phase to the platform vibration; simultaneously adjusting the command amplitude according to the real-time state error in step 43 to match the vibration amplitude (e.g., when the vibration acceleration is 0.1g, the actuator output corresponding to the command meets the force value required to counteract the acceleration); and finally, the output command format is adapted to the actuator's interface requirements, allowing direct driving of the actuator array to generate a reverse force.

[0037] In this embodiment of the invention, initializing the adaptive controller parameters and the recursive least squares covariance matrix allows the controller to start from a baseline state adapted to the characteristics of the 0.25m camera platform, ensuring computational stability. Using the corrected vibration state vector as input, which has eliminated nonlinear deviations, provides the controller with a precise initial state basis. The feedforward control channel, based on a preset platform vibration model (covering typical scenarios such as thermal deformation and resonance), can pre-calculate control components to address known vibrations, reducing response delays when vibrations occur. The model closely matches the dynamic characteristics of the camera's rigid structure, making the feedforward components more adaptable to actual vibration patterns and improving the foresight of active cancellation. Calculating real-time state errors accurately captures the deviation between the current state and the stable target. The recursive least squares method updates the control law parameters in real time. The system adapts to the dynamic drift caused by thermal deformation and fuel consumption of the platform, ensuring that the feedback component is dynamically adjusted according to the operating conditions; the feedback channel can specifically correct residual deviations not covered by the feedforward, improving control accuracy; by fusing the feedforward and feedback components and weighting and summing them, the system can integrate the advantages of both (feedforward handles known vibrations, and feedback corrects real-time deviations); amplitude limiting processing can prevent commands from exceeding the output range of the actuator (piezoelectric ceramic / electromagnetic actuator), preventing actuator overload and ensuring the safety and feasibility of control commands; output conversion of preliminary commands can make the command format adaptable to the actuator drive requirements (such as the voltage signal of piezoelectric ceramic and the current signal of electromagnetic actuator); ensuring that the final command is out of phase and matches the vibration, it can accurately drive the actuator to generate a reverse force, which meets the vibration cancellation requirements of camera optical components.

[0038] In a preferred embodiment of the present invention, step 5 above, which involves driving a piezoelectric ceramic actuator or an electromagnetic actuator array according to the reverse control command to generate a reverse force with opposite phase and matching amplitude to the vibration, and applying this reverse force to the imaging optical component to obtain a change in the platform vibration state, includes: Step 51: Based on the amplitude and phase parameters of the reverse control command, obtain the corresponding actuator drive signal. Specifically, this includes: extracting the amplitude and phase parameters from the reverse control command; for piezoelectric ceramic actuators, converting the command amplitude into the duty cycle of the corresponding PWM drive signal (matching its response speed of less than 1ms and output density of greater than 100N / kg), and converting the phase parameters into signal trigger timing to ensure that the drive signal is in the opposite phase to the vibration; for electromagnetic actuator arrays, converting the command amplitude into the magnitude of a constant current signal (set according to its rated output range), and converting the phase parameters into current direction switching timing; the drive signal format is also adapted to the output characteristics of the secondary power supply (overall efficiency greater than or equal to 75%) to ensure no significant attenuation during signal transmission, ultimately obtaining an actuator drive signal that precisely corresponds to the command parameters.

[0039] Step 52: Based on the actuator drive signal, drive the piezoelectric ceramic actuator or electromagnetic actuator array to generate mechanical force that is opposite in phase and matches the amplitude of the platform vibration. Specifically, this includes: based on the actuator drive signal from step 51, when driving the actuator, maintain the operating temperature of the piezoelectric ceramic actuator or electromagnetic actuator within the range of -20°C to 55°C of the cabin electronic equipment using the camera thermal control component to avoid temperature drift affecting the output accuracy; for the piezoelectric ceramic actuator, control its extension and contraction through the drive signal to generate mechanical force that is opposite in phase to the platform vibration, and match the output amplitude to the vibration amplitude as required by the instruction (e.g., when the vibration acceleration is 0.1g, the corresponding output force meets the force value required to counteract the acceleration); for the electromagnetic actuator array, control the coil current through the drive signal to generate a uniformly distributed thrust to ensure that the overall output of the array is opposite in phase and matches the amplitude of the vibration; the entire driving process refers to the bandwidth requirement of 0.1 to 200Hz of the dynamic resonance suppression and stabilization system to ensure that the mechanical force can cover the main vibration frequency and specifically counteract the vibration.

[0040] Step 53 involves applying the mechanical force as a counterforce to the mounting base of the imaging optical component or an active vibration isolation mechanism. This mechanical coupling counteracts platform vibration, allowing for the acquisition of changes in the platform's vibration state. Specifically, this includes: applying the mechanical force as a counterforce to the mounting base of the imaging optical component (utilizing the rigid structure of the camera's main load-bearing frame to ensure efficient force transmission) or an active vibration isolation mechanism (compliant with the structural design of a dynamic resonance suppression and image stabilization system); adhering to the installation accuracy requirements of the optomechanical interface (e.g., a coplanarity of 0.1 mm on the mounting surface) during application to prevent the force from causing positional shifts in the optical component; and through the rigid mechanical coupling between the optical component and the platform, allowing the counterforce to act directly on the key vibration carrier, efficiently counteracting platform vibration in the three degrees of freedom. Simultaneously, because the force is directly related to the vibration state of the optical component, changes in the platform's vibration state (e.g., the attenuation of displacement, velocity, and angular velocity) can be acquired in real-time and accurately.

[0041] In this embodiment of the invention, an actuator drive signal is generated based on the amplitude and phase parameters of the reverse control command. This ensures that the drive signal accurately corresponds to the mechanical parameter requirements in the command, adapting to the drive characteristics of piezoelectric ceramic actuators (response speed less than 1ms) or electromagnetic actuators. Simultaneously, the signal format matches the actuator interface requirements (e.g., PWM drive for piezoelectric ceramics, current drive for electromagnetic actuators), providing a suitable signal foundation for stable actuator output and avoiding output deviations caused by signal mismatch. Based on the drive signal, the actuator array is driven, leveraging the characteristics of the actuators (piezoelectric ceramic output density greater than 100N / kg, electromagnetic actuator rated output stable) to accurately generate a mechanical vibration with opposite phase and matching amplitude to the platform vibration. Mechanical power is applied during the output process to meet the 0.1 to 200 Hz bandwidth requirements of the dynamic resonance suppression and stabilization system, ensuring that the force can specifically counteract the current vibration frequency components and meet the real-time and accuracy requirements of the camera optical components for vibration suppression. Applying mechanical power to the mounting base or active vibration isolation mechanism of the imaging optical components can efficiently transmit the reverse force to counteract vibration by means of the rigid mechanical coupling between the optical components and the platform (such as the stable support of the main load-bearing frame). At the same time, the force acts directly on the key vibration carrier (optical components), which can accurately obtain the changes in the platform's vibration state. This change data can truly reflect the vibration suppression effect, providing a clear state basis for subsequent feedback signal acquisition and ensuring the continuity of closed-loop control.

[0042] In a preferred embodiment of the present invention, step 6 above, based on the change in platform vibration state, involves real-time acquisition of the platform response signal after vibration cancellation via the sensor array, including: Step 61: Based on the changes in the platform's vibration state, a synchronous acquisition trigger command for the sensor array is obtained. Specifically, this includes: generating a synchronous acquisition trigger command for the sensor array based on changes in the platform's vibration state (such as real-time dynamic changes in three-degree-of-freedom displacement decay and angular velocity reduction), combined with the actuator output timing (the piezoelectric ceramic actuator response speed is less than 1ms, and the electromagnetic actuator array thrust output cycle); the generation frequency of the trigger command matches the 8Hz integral time update frequency of the focal plane circuit, and is linked with the GNSS second pulse signal (RS422 level specification, triggered at the leading edge of the second) to ensure that the acquisition timing accurately corresponds to the vibration response stage after the application of the reverse force; the command format is adapted to the control interface requirements of the sensor array (compatible with the 500kbps rate of the focal plane circuit CAN bus) to avoid timing misalignment causing the acquired response signal to become disconnected from the vibration suppression effect.

[0043] Step 62: In response to the synchronous acquisition trigger command, control the multi-axis accelerometer and gyroscope to work synchronously, and acquire the three-degree-of-freedom vibration acceleration and angular velocity response signals generated by the imaging platform after the application of the reverse force, which characterize the vibration control effect. Specifically, this includes: when responding to the synchronous acquisition trigger command, activating the three-axis MEMS accelerometer (0.1mg-level accuracy) and fiber optic gyroscope (zero-bias stability less than or equal to 0.01° / h) through the sensor array control unit, and using GNSS second pulse signals to achieve synchronous operation of the two, ensuring that the time deviation of a single sampling does not exceed 20μs; during the acquisition process, relying on the camera thermal control component, maintaining the sensor operating environment temperature within the range of -20℃ to 55℃ of the cabin electronic equipment to reduce the impact of temperature drift on measurement accuracy; the acquired response signals cover the three degrees of freedom of the platform, including vibration acceleration in the X, Y, and Z axes (reflecting the linear vibration cancellation effect) and angular velocity around the three axes (reflecting the rotational vibration cancellation effect), with a signal sampling frequency of not less than 500Hz to fully capture the dynamic changes within the 0.1 to 200Hz vibration bandwidth, ensuring that the acquired data can directly characterize the vibration control effect of the reverse force.

[0044] Step 63: Preprocess the vibration acceleration and angular velocity response signals to obtain digital signals with a uniform format. Specifically, this includes: preprocessing the acquired vibration acceleration and angular velocity response signals; using a sliding window method to remove pulse-like anomalies caused by instantaneous sensor interference; and formatting the acceleration (m / s²) signals into a uniform digital signal format. 2 Signals of magnitude (on the order of magnitude) and angular velocity (on the order of rad / s) are uniformly converted into 16-bit digital signal format. During the conversion process, the mean and standard deviation of historical vibration data are used to eliminate dimensional differences, so that signals of different dimensions are on the same numerical scale. At the same time, the signal is digitally filtered, and the filtering bandwidth matches the requirements of the dynamic resonance suppression and stabilization system from 0.1 to 200 Hz, filtering out high-frequency noise that exceeds the vibration frequency range, and finally obtaining a digital signal with uniform format and noise suppression.

[0045] Step 64: Based on the standardized digital signal, analyze and calculate the displacement, attitude angle change rate, and motion trajectory curvature characteristics of the platform in three degrees of freedom, to obtain a response signal containing the platform's motion geometric feature information. Specifically, this includes: based on the standardized digital signal, according to the 0.25m camera Cartesian right-hand coordinate system (+Z direction is the optical axis), converting the vibration acceleration signal into three-degree-of-freedom displacement parameters (X, Y, Z directions) through integration, and converting the angular velocity signal into attitude angle change rate (around the X, Y, Z axes); further combining the displacement and attitude angle change rate data, calculating the curvature characteristics of the platform's motion trajectory (reflecting the degree of curvature of the vibration trajectory, adapting to the needs of complex vibration mode sensing); in the calculation process, inertial parameters of the camera's optical engine main body design (such as the main body weight of approximately 290kg and rotational inertia) are incorporated to correct for accumulated integration errors and ensure the calculation accuracy of displacement, attitude angle change rate, and curvature characteristics; finally, a response signal containing the above geometric feature information is obtained.

[0046] Step 65 involves encapsulating and aligning the response signal containing the platform's motion geometric features to obtain the vibration-cancelled platform response signal. Specifically, this includes: encapsulating the response signal containing the platform's motion geometric features, including displacement, attitude angle change rate, trajectory curvature, and simultaneously incorporating the signal acquisition timestamp (based on GNSS second pulse generation, with an accuracy better than 0.05ms); aligning the encapsulated signal format with the focal plane circuit data output format (16-bit digital signal, CAN bus compatible frame structure) according to the requirements of the adaptive controller feedback interface based on recursive least squares method, ensuring that the geometric feature parameters of each dimension are accurately matched with the controller parameter tuning logic, and finally obtaining the vibration-cancelled platform response signal.

[0047] In this embodiment of the invention, a synchronous acquisition trigger command for the sensor array is generated based on the changes in the platform's vibration state. This ensures that the acquisition timing matches the vibration response stage after the application of the reverse force, avoiding invalid data caused by premature or delayed acquisition. The trigger command is synchronized with the actuator output timing to ensure that the acquired response signal can be directly correlated with the vibration suppression effect. The synchronous acquisition trigger command controls the multi-axis accelerometer (0.1mg-level accuracy) and gyroscope (zero-bias stability less than or equal to 0.01° / h) to work synchronously, eliminating timing deviations between sensors and ensuring the time consistency of the three-degree-of-freedom vibration acceleration and angular velocity response signals. The acquired signal directly characterizes the vibration control effect of the reverse force, fully capturing the dynamic changes after vibration cancellation, and the signal accuracy meets the camera's accuracy requirements for vibration monitoring. The vibration acceleration and angular velocity response signals are preprocessed and converted into a uniform digital signal format according to the 16-bit digital signal output format of the focal plane circuit, eliminating inconsistencies. The preprocessing process addresses the differences in dimensions and signal scales between the sensor outputs; it also removes transient interference noise from the signal to ensure the stability of the digital signal; based on a digital signal analysis and calculation platform with a unified format, it can deeply mine the deep geometric motion information contained in the vibration response, meeting the camera's need for sensing complex vibration modes; the extracted geometric feature information can intuitively reflect the motion law after vibration cancellation, so that the response signal not only contains basic vibration parameters, but also key information characterizing the platform's motion characteristics, providing a richer basis for controller parameter tuning; the response signal containing geometric feature information is encapsulated and format-aligned to meet the feedback interface requirements of the adaptive controller (consistent with the data transmission format of the focal plane circuit); format alignment ensures that the parameters of each dimension of the signal (displacement, attitude angle change rate, etc.) can be accurately matched with the controller parameter tuning logic, avoiding information loss due to format deviation.

[0048] In a preferred embodiment of the present invention, step 7 above, utilizing the platform response signal after vibration cancellation, uses a closed-loop feedback mechanism to tune the parameters of the adaptive controller online, thereby achieving continuous optimization of the vibration suppression effect, including: Step 71: Extract the geometric feature information and dynamic response parameters contained in the vibration-cancelled platform response signal to obtain the vibration suppression effect evaluation index. Specifically, this includes: extracting the geometric feature information contained in the vibration-cancelled platform response signal, including the three-degree-of-freedom displacement (X, Y, Z directions), attitude angle change rate (around the X, Y, Z axes), and motion trajectory curvature resolved based on the camera's Cartesian right-hand coordinate system (+Z direction is the optical axis direction); simultaneously extracting the dynamic response parameters, covering the vibration frequency (0.1 to 200 Hz, matching the bandwidth of the dynamic resonance suppression stabilization system), amplitude attenuation rate, and response delay time; integrating this information into the vibration suppression effect evaluation index, and simultaneously incorporating the measurement accuracy thresholds of the three-axis MEMS accelerometer (0.1 mg level accuracy) and fiber optic gyroscope (zero bias stability less than or equal to 0.01° / h) to ensure that the index can accurately quantify the actual impact of vibration suppression on imaging.

[0049] Step 72: Based on the vibration suppression effect evaluation index, calculate the performance deviation between the current control effect and the desired stable state. Specifically, this includes: based on the vibration suppression effect evaluation index, taking the desired stable state when the camera is imaging at 0.25m as the benchmark, in which the platform's three-degree-of-freedom displacement approaches 0, the attitude angle change rate approaches 0, and the vibration frequency and amplitude are within the allowable range of the dynamic resonance suppression system (e.g., amplitude less than or equal to 5mg, matching the platform's micro-vibration control requirements); calculating the difference between the current control effect and this benchmark in each index dimension to obtain the performance deviation; during the deviation calculation process, the influence of the camera's thermal control environment on the vibration characteristics is considered to correct the index deviation caused by temperature drift, ensuring that the performance deviation can truly reflect the gap between the control effect and the ideal state.

[0050] Step 73: Based on the performance deviation, adjust the weighting coefficients and control law parameters of the feedforward and feedback control channels online using the recursive least squares method to obtain the control parameter adjustment amount. Specifically, this includes: based on the performance deviation in step 72, calling the recursive least squares method to perform online parameter adjustment; when adjusting the weighting coefficients of the feedforward control channel, if the deviation originates from periodic vibration (such as CMG maneuvering), increase the feedforward weight (0.6 to 0.8), and if it originates from random vibration, decrease the feedforward weight (0.2 to 0.4); when adjusting the weighting coefficients of the feedback control channel, dynamically adapt according to the magnitude of the deviation, and increase the feedback weight (0.6 to 0.8) when the deviation is large to quickly correct it; synchronously adjust the control law parameters, incorporating platform dynamic drift factors (such as changes in structural stiffness caused by thermal deformation, and fine-tuning of the 290kg weight distribution of the main body), and synchronizing the parameter update frequency with the camera control cycle (8Hz, matching the focal plane circuit integration time update frequency); during the adjustment process, ensure that the power consumption meets the requirements, does not exceed the long-term average thermal control power consumption budget of 360W for the camera, and adapts to a secondary power supply efficiency of greater than or equal to 75% of the overall system efficiency.

[0051] Step 74 involves updating the control parameter adjustment amount to the adaptive controller, completing the online tuning of the controller parameters, and obtaining the adaptive controller with updated parameters. Specifically, this includes: transmitting the control parameter adjustment amount obtained in step 73 to the adaptive controller via a CAN bus (rate 500kbps, meeting camera communication interface requirements); using a 16-bit digital signal format consistent with the focal plane circuit during transmission to ensure that the parameter format is compatible with the controller input interface; following a gradual adjustment principle during updates to avoid parameter abrupt changes causing actuator (such as piezoelectric ceramic actuators with a force density greater than 100N / kg) overload; verifying whether the adjusted parameters are within the safe operating range of the controller and actuator; and immediately loading the new parameters after the update without interrupting the current vibration suppression process to ensure control continuity.

[0052] Step 75: Based on the updated parameters of the adaptive controller, the vibration suppression operation for the next control cycle is executed to continuously optimize the vibration suppression effect. Specifically, this includes: based on the updated parameters of the adaptive controller, the vibration suppression operation for the next control cycle is executed according to an 8Hz control cycle (matching the update frequency of the focal plane circuit integral time); relying on the computing power of the camera's intelligent supercomputing-driven image depth analysis engine, real-time vibration signals are first collected through the sensor array in each cycle, and then the processing logic from Steps 2 to 6 (Kalman filtering noise reduction, deep neural network correction, feedforward and feedback fusion) is used to generate control commands; during the process, the response signals from the previous cycle are continuously used to optimize the parameters to adapt to the dynamic characteristic drift of the platform caused by thermal deformation (cabin temperature -20℃ to 55℃) and fuel consumption, ensuring that the vibration suppression effect is continuously optimized as the mission progresses.

[0053] In this embodiment of the invention, extracting geometric feature information (such as displacement, attitude angle change rate, and trajectory curvature) and dynamic response parameters from the response signal allows the vibration suppression effect evaluation index to better match the motion characteristics of the camera optical components. The index covers deep motion characteristics, reflecting the suppression effect more comprehensively than shallow parameters, providing a precise evaluation basis for subsequent parameter adjustments. Calculating performance deviation based on the evaluation index directly quantifies the gap between the current control effect and the desired stable state. Deviation calculation, combined with the camera's dynamic resonance suppression requirements, ensures that the deviation value accurately points to weak control links. Online adjustment of control parameters using the recursive least squares method allows for real-time dynamic optimization of the weighting coefficients and control law of the feedforward and feedback channels using response signal data. This method is adaptable to platforms experiencing thermal deformation and structural vibration. The dynamic drift of the launch vehicle does not rely on ground calibration models, allowing parameter adjustments to fit the real-time on-orbit operating conditions, ensuring adjustment accuracy and timeliness. The control parameter adjustments are updated to the adaptive controller in real time, enabling the controller to quickly adapt to the current vibration state. The update process is synchronized with the camera control cycle (matching the 8Hz integral time update frequency of the focal plane circuit), without interrupting the vibration suppression process, ensuring that the controller parameters match the actual needs in real time. The controller executes the next cycle operation based on the updated parameters, forming a closed-loop optimization of acquisition, evaluation, adjustment, and execution. Each cycle optimizes parameters based on the response data of the previous cycle, continuously adapting to changes in platform vibration modes (such as resonant frequency shifts and changes in vibration characteristics caused by thermal deformation), ensuring that the vibration suppression effect is continuously optimized as the mission progresses.

[0054] like Figure 2 As shown, embodiments of the present invention also provide a dynamic resonance suppression and image stabilization feedback control system, comprising: The acquisition module is used to acquire the vibration acceleration and angular velocity signals of the platform in three degrees of freedom in real time through a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. The fusion module is used to perform noise reduction on the vibration acceleration and angular velocity signals using the Kalman filter algorithm to obtain the noise-reduced platform state signal; the noise-reduced platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. The correction module is used to input the vibration state vector into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount. The control module is used to input the corrected vibration state vector into the adaptive controller based on the recursive least squares method, and generate inverse control commands to counteract the vibration in real time through a combination of feedforward and feedback. The execution module is used to drive the piezoelectric ceramic actuator or electromagnetic actuator array according to the reverse control command to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and apply the reverse force to the imaging optical component to obtain the change in the vibration state of the platform. The post-cancellation acquisition module is used to acquire the platform response signal after vibration cancellation in real time through the sensor array based on the changes in the platform's vibration state. The closed-loop optimization module is used to utilize the platform response signal after vibration cancellation to tune the parameters of the adaptive controller online through a closed-loop feedback mechanism, so as to achieve continuous optimization of the vibration suppression effect.

[0055] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0056] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0057] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0058] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A dynamic resonance suppression and image stabilization feedback control method, characterized in that, The method includes: The vibration acceleration and angular velocity signals of the platform in three degrees of freedom are collected in real time by a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. The Kalman filter algorithm is used to denoise the vibration acceleration and angular velocity signals to obtain the denoised platform state signal. The denoised platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. The vibration state vector is input into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount. The corrected vibration state vector is input into an adaptive controller based on recursive least squares, and inverse control commands to counteract vibration are generated in real time through a combination of feedforward and feedback. According to the reverse control command, the piezoelectric ceramic actuator or electromagnetic actuator array is driven to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and the reverse force is applied to the imaging optical component to obtain the change in the platform vibration state. Based on the changes in the platform's vibration state, the platform's response signal after vibration cancellation is collected in real time through the sensor array. By utilizing the platform response signal after vibration cancellation, the parameters of the adaptive controller are tuned online through a closed-loop feedback mechanism to achieve continuous optimization of the vibration suppression effect.

2. The dynamic resonance suppression and image stabilization feedback control method according to claim 1, characterized in that, The Kalman filter algorithm is used to denoise the vibration acceleration and angular velocity signals to obtain the denoised platform state signal. The denoised platform state signal is then fused using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform, including: The vibration acceleration and angular velocity signals are preprocessed to obtain preprocessed signals; based on the preprocessed signals, a Kalman filter state equation describing the dynamic characteristics is established; according to the state equation and the sensor measurement principle, the Kalman filter observation equation is constructed. Based on the state equation and observation equation, the state prediction stage uses the state estimate and its covariance matrix from the previous moment, combined with the platform dynamics characteristics described by the state equation, to calculate and predict the platform state vector and covariance matrix at the current moment. Based on the platform state vector and covariance matrix, the Kalman gain matrix is ​​calculated during the observation update phase. The predicted state and covariance matrix are then corrected using the gain matrix and the actual observation value at the current time to obtain the final state estimate and the updated covariance matrix, thus obtaining the denoised platform state signal. The platform state signal after noise reduction is used to perform multi-source information fusion and attitude calculation using a quaternion-based complementary filtering algorithm, and the real-time vibration state vector of the platform is directly output.

3. The dynamic resonance suppression and image stabilization feedback control method according to claim 2, characterized in that, The vibration state vector is input into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction value, including: Obtain a pre-trained deep neural network model, which is used to establish a nonlinear mapping relationship from vibration state to its correction amount; The real-time vibration state vector of the platform is standardized and preprocessed to obtain a standardized vector that conforms to the input format of the deep neural network model. The standardized vector is input into the deep neural network model, and the corresponding vibration state correction vector is obtained through multi-layer nonlinear transformation calculations within the model. Based on the vibration state correction vector, the real-time vibration state vector of the platform is compensated and corrected to obtain the corrected vibration state vector.

4. The dynamic resonance suppression and image stabilization feedback control method according to claim 3, characterized in that, The corrected vibration state vector is input to an adaptive controller based on recursive least squares, which generates inverse control commands in real time to counteract the vibration through a combination of feedforward and feedback, including: Initialize the control parameters of the adaptive controller and the covariance matrix required by the recursive least squares method; use the corrected vibration state vector as the state input for the current control cycle; Based on the state input, the feedforward control components are calculated by forward derivation according to the pre-set platform vibration model through the feedforward control channel. Based on the state input, the difference between it and the desired steady state is calculated to obtain the real-time state error; based on the real-time state error, the control law parameters are updated in real time using the recursive least squares method through the feedback control channel, and the feedback control components are calculated. By fusing the feedforward control component and the feedback control component, performing weighted summation and amplitude limiting processing, a preliminary reverse control command is obtained. The initial inverse control command is output-converted to obtain the final inverse control command that is opposite in phase and matches the amplitude of the vibration, used to drive the actuator array.

5. The dynamic resonance suppression and image stabilization feedback control method according to claim 4, characterized in that, According to the reverse control command, a piezoelectric ceramic actuator or electromagnetic actuator array is driven to generate a reverse force with opposite phase and matching amplitude to the vibration, and this reverse force is applied to the imaging optical components to obtain changes in the platform vibration state, including: Based on the amplitude and phase parameters of the reverse control command, the corresponding actuator drive signal is obtained; Based on the actuator drive signal, the piezoelectric ceramic actuator or electromagnetic actuator array is driven to generate mechanical force that is opposite in phase and matches the amplitude of the platform vibration. The mechanical force is used as a reverse force and applied to the mounting base of the imaging optical component or the active vibration isolation mechanism. Through mechanical coupling, the platform vibration is counteracted, and the change in the platform vibration state is obtained.

6. The dynamic resonance suppression and image stabilization feedback control method according to claim 5, characterized in that, Based on changes in platform vibration state, the platform response signal after vibration cancellation is acquired in real time through the sensor array, including: Based on the changes in the vibration state of the platform, a synchronous acquisition trigger command for the sensor array is obtained; In response to the synchronous acquisition trigger command, the multi-axis accelerometer and gyroscope are controlled to work synchronously to acquire the three-degree-of-freedom vibration acceleration and angular velocity response signals generated by the imaging platform after the application of the reverse force, which characterize the vibration control effect; The vibration acceleration and angular velocity response signals are preprocessed to obtain digital signals with a uniform format; Based on the standardized digital signal, the displacement, attitude angle change rate and motion trajectory curvature characteristics of the platform in the three degrees of freedom are analyzed and calculated to obtain a response signal containing the platform's motion geometric feature information. The response signal containing the platform motion geometry information is encapsulated and format-aligned to obtain the vibration-cancelled platform response signal.

7. The dynamic resonance suppression and image stabilization feedback control method according to claim 6, characterized in that, Using the platform response signal after vibration cancellation, the parameters of the adaptive controller are tuned online through a closed-loop feedback mechanism to continuously optimize the vibration suppression effect, including: Geometric feature information and dynamic response parameters contained in the platform response signal after vibration cancellation are extracted to obtain vibration suppression effect evaluation index; Based on the vibration suppression effect evaluation index, calculate the performance deviation between the current control effect and the desired steady state; Based on the performance deviation, the weighting coefficients and control law parameters of the feedforward and feedback control channels are adjusted online using the recursive least squares method to obtain the control parameter adjustment amount; The control parameter adjustment amount is updated to the adaptive controller to complete the online tuning of the controller parameters and obtain the adaptive controller with updated parameters; Based on the updated parameters of the adaptive controller, the vibration suppression operation is executed in the next control cycle to achieve continuous optimization of the vibration suppression effect.

8. A dynamic resonance suppression and image stabilization feedback control system, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire the vibration acceleration and angular velocity signals of the platform in three degrees of freedom in real time through a sensor array consisting of multi-axis accelerometers and gyroscopes deployed on the imaging platform. The fusion module is used to perform noise reduction on the vibration acceleration and angular velocity signals using the Kalman filter algorithm to obtain the noise-reduced platform state signal; the noise-reduced platform state signal is then fused with multi-source information using a sensor fusion algorithm to obtain the real-time vibration state vector of the platform. The correction module is used to input the vibration state vector into a pre-trained deep neural network model, which directly outputs the corrected vibration state vector. The deep neural network model is trained using historical vibration data and is used to establish a nonlinear mapping relationship from the vibration state to its correction amount. The control module is used to input the corrected vibration state vector into the adaptive controller based on the recursive least squares method, and generate inverse control commands to counteract the vibration in real time through a combination of feedforward and feedback. The execution module is used to drive the piezoelectric ceramic actuator or electromagnetic actuator array according to the reverse control command to generate a reverse force that is opposite in phase and matches the amplitude of the vibration, and apply the reverse force to the imaging optical component to obtain the change in the vibration state of the platform. The post-cancellation acquisition module is used to acquire the platform response signal after vibration cancellation in real time through the sensor array based on the changes in the platform's vibration state. The closed-loop optimization module is used to utilize the platform response signal after vibration cancellation to tune the parameters of the adaptive controller online through a closed-loop feedback mechanism, so as to achieve continuous optimization of the vibration suppression effect.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images