A Laser Beam Pointing Stabilization Control Method and System Based on LSTM-PID

By using an LSTM-PID composite control architecture, combining LSTM timing feedforward and PID closed-loop feedback, stable control of the laser beam pointing was achieved, solving the problems of response lag and long-term drift in existing technologies, and improving the stability and accuracy of the laser system.

CN123113341APending Publication Date: 2026-11-17SHANXI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611298837.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing laser beam pointing stabilization control methods suffer from problems such as response lag, insufficient ability to suppress complex disturbances, easy drift during long-term operation, and ineffective coordination between prediction feedforward and closed-loop feedback, making it difficult to meet the usage requirements of high-power, high-precision, and long-duration laser systems.

Method used

An LSTM-PID composite control architecture is adopted. By combining LSTM time-series feedforward prediction with PID real-time closed-loop feedback, a multi-dimensional time-series dataset is constructed using the spot position and control voltage signal. An LSTM prediction model is trained to predict the spot offset, and then combined with PID closed-loop control to generate a weighted and fused control voltage signal to drive the beam deflection actuator.

Benefits of technology

It achieves advance compensation for beam offset and real-time correction of instantaneous errors, improves the stability of laser beams in complex interference environments, balances dynamic response speed and long-term steady-state accuracy, and significantly reduces beam offset errors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113341A_ABST
    Figure CN123113341A_ABST
Patent Text Reader

Abstract

This application discloses a laser beam pointing stabilization control method and system based on LSTM-PID, relating to the fields of laser precision control and optical stabilization technology. The method includes acquiring spot position signals and control voltage signals to construct a multi-dimensional time-series dataset; training an LSTM prediction model; acquiring historical multi-dimensional time-series data; using the trained LSTM prediction model to predict the spot offset at the next moment, generating a feedforward control quantity; calculating a real-time feedback control quantity through PID closed-loop control based on the deviation between the current actual spot position and the target zero-point position; and weightedly fusing the feedforward control quantity and the real-time feedback control quantity to output the final control voltage signal to compensate for beam pointing errors and achieve laser beam pointing stabilization. This application can achieve early beam offset compensation and real-time instantaneous error correction, balancing dynamic response speed and long-term steady-state accuracy, and improving the laser beam pointing stability under complex interference environments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of laser precision control and optical stabilization technology, and in particular to a laser beam pointing stabilization control method and system based on LSTM-PID. Background Technology

[0002] Laser beam pointing stability is a key technical indicator for various high-precision laser systems. Beam pointing jitter and drift directly reduce the system's positioning accuracy, energy utilization efficiency, and long-term operational reliability. During free-space propagation, laser beams are subject to various internal and external interferences, such as mechanical vibration, air turbulence disturbances, ambient temperature drift, and optical platform deformation, causing random high-frequency jitter and slow low-frequency drift in the beam spot.

[0003] Most existing beam pointing stabilization control methods employ traditional PID (Proportional-Integral-Derivative) closed-loop feedback control. This involves using a position-sensitive detector to collect the current position deviation of the beam spot and then driving the actuator to complete the correction. While this method is simple to implement in hardware, PID is a pure time-delay feedback control, relying solely on the current error for adjustment and unable to predict the subsequent beam deviation trend. When faced with nonlinear, strongly coupled, and long-term environmental disturbances, the system's correction response exhibits significant lag; zero-point drift is difficult to eliminate under long-term continuous operation, and the steady-state control accuracy has an inherent upper limit.

[0004] Existing improvement methods include adaptive control, fuzzy control, and model predictive control. However, most of these algorithms heavily rely on accurate physical models of the system, making parameter tuning difficult and engineering implementation complex. While temporal deep learning networks such as LSTM (Long Short-Term Memory) can capture the long-term temporal dependencies of disturbances and predict offset trends, their closed-loop stability is poor when used alone as a controller, easily leading to overshoot and oscillations, making them difficult to apply directly to high-precision optical closed-loop control systems.

[0005] In summary, existing beam pointing stabilization technologies suffer from drawbacks such as response lag, insufficient ability to suppress complex disturbances, significant drift over long periods of operation, and ineffective coordination between prediction feedforward and closed-loop feedback, making it difficult to meet the engineering requirements of high-power, high-precision, and long-duration laser systems. Summary of the Invention

[0006] The purpose of this application is to provide a laser beam pointing stability control method and system based on LSTM-PID. Through the composite control architecture of LSTM timing feedforward prediction and PID real-time closed-loop feedback, it can realize beam offset advance compensation and instantaneous error real-time correction, taking into account dynamic response speed and long-term steady-state accuracy, and improving the pointing stability of laser beams under complex interference environments.

[0007] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a laser beam pointing stabilization control method based on LSTM-PID, comprising the following steps: Acquire the spot position signal and control voltage signal.

[0008] A multidimensional time-series dataset is constructed based on the spot position signal and the control voltage signal.

[0009] Based on the multidimensional time series dataset, the LSTM prediction model is trained to obtain a trained LSTM prediction model.

[0010] The historical multidimensional time series data obtained by the sliding window is acquired, and based on the historical multidimensional time series data, the trained LSTM prediction model is used to predict the spot offset at the next moment and generate the corresponding feedforward control quantity; the historical multidimensional time series data includes the current actual position of the spot and the control voltage at the corresponding moment.

[0011] Based on the deviation between the current actual position of the light spot and the target zero point position, the corresponding real-time feedback control quantity is calculated through PID closed-loop control.

[0012] The feedforward control quantity and the real-time feedback control quantity are weighted and fused to output a final control voltage signal; the final control voltage signal is used to control the beam deflection actuator to deflect at an angle to compensate for the beam pointing error and achieve laser beam pointing stability.

[0013] Secondly, this application provides a laser beam pointing stabilization control system based on LSTM-PID, the system being used to implement the laser beam pointing stabilization control method based on LSTM-PID, the system comprising an optical path unit, a detection unit, a control unit, and an execution unit.

[0014] The optical path unit is used to generate a laser beam and collimate, transmit, split, and adjust the power of the beam so that the power of the beam incident on the detection unit is within the linear operating range.

[0015] The detection unit is used to acquire the position signal of the far-field light spot in real time and convert the position signal into an electrical signal and output it to the control unit.

[0016] The control unit includes an LSTM timing prediction module and a PID closed-loop control module. The LSTM timing prediction module is used to predict the spot offset at the next moment based on historical multi-dimensional timing data and generate a corresponding feedforward control quantity. The PID closed-loop control module is used to calculate the corresponding real-time feedback control quantity based on the deviation between the current actual spot position and the target zero point position. The control unit is also used to perform weighted fusion of the feedforward control quantity and the real-time feedback control quantity, output the final control voltage signal, and send it to the execution unit.

[0017] The execution unit is a beam deflection actuator, which is used to generate an angle deflection according to the final control voltage signal to compensate for beam pointing error and achieve laser beam pointing stability.

[0018] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a laser beam pointing stabilization control method and system based on LSTM-PID. It utilizes a multi-dimensional time-series dataset containing the beam spot position and corresponding control voltage to train an LSTM prediction model. By mining the relevant variation patterns of beam spot offset based on the time-series data, it provides a data foundation for beam spot offset prediction and supports the effective generation of feedforward control quantities. By learning from historical multi-dimensional time-series data captured by the sliding window using the LSTM prediction model, it can predict the beam spot offset at the next moment based on historical time-series information and generate feedforward control quantities. This allows for the prediction of beam spot offset development trends, forming feedforward compensation and improving the adjustment lag problem caused by simply relying on PID closed-loop feedback and adjusting only based on the current deviation. Combined with PID closed-loop control, the real-time feedback control quantity is calculated based on the deviation between the current actual beam spot position and the target zero-point position, and the actual beam pointing error at the current moment is corrected in real time, ensuring the closed-loop correction capability of the control. The final control voltage signal is obtained by weighted fusion of the feedforward control quantity output by LSTM and the real-time feedback control quantity output by PID. This achieves effective fusion of LSTM timing feedforward prediction and PID real-time closed-loop feedback. The LSTM feedforward prediction compensation and PID closed-loop feedback correction work together to drive the beam deflection actuator to deflect the beam angle to compensate for the beam pointing error. This can achieve early compensation for beam offset and real-time correction of instantaneous error, taking into account both dynamic response speed and long-term steady-state accuracy. It can achieve precise and stable control of laser beam pointing and improve the stability of laser beam pointing under complex interference environments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic flowchart of a laser beam pointing stabilization control method based on LSTM-PID provided in an embodiment of this application; Figure 2 A schematic diagram of the overall structure of a laser beam pointing stabilization control system based on LSTM-PID provided in an embodiment of this application; Figure 3 This is a schematic diagram of the detector's photosensitivity response curve provided in an embodiment of this application; Figure 4 This is a schematic diagram of the effective detection area of ​​the detector provided in one embodiment of this application; Figure 5 This is a schematic diagram illustrating the geometric relationship between the feedback frame deflection angle and the measurement distance, provided in an embodiment of this application. Figure 6 This is a schematic diagram of the LSTM prediction model structure provided in an embodiment of this application; Figure 7 This is a schematic diagram of a position during 3 hours of free-running without feedback, provided as an embodiment of this application. Figure 8 This is a schematic diagram showing the position after 3 hours of operation with LSTM-PID feedback, provided in one embodiment of this application.

[0021] Reference numerals: 1-Laser, 2-Focusing lens, 3-First high-reflectivity mirror, 4-Second high-reflectivity mirror, 5-Half-wave plate, 6-Beam splitter, 7-Power meter, 8-First feedback mount, 9-Second feedback mount, 10-Third high-reflectivity mirror, 11-Fourth high-reflectivity mirror, 12-First position-sensitive detector, 13-Beam collector, 14-Second position-sensitive detector, 15-Beam positioning controller, 16-Piezoelectric controller, 17-Computer. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] This application provides a laser beam pointing stabilization control method and system based on LSTM-PID, which can be applied to scenarios requiring long-term high pointing stability of laser beams, such as space optical communication, quantum precision measurement, high-precision laser processing, and precision optical experiments. By combining LSTM timing prediction feedforward with PID closed-loop feedback, it achieves advanced suppression and real-time correction of beam pointing jitter, significantly reducing beam offset error and improving the stability of the system under long-term, strong interference. It meets the requirements of high-precision laser systems for beam pointing control and solves the problems of response lag, weak nonlinear disturbance suppression capability, easy drift during long-term operation, and inability of pure PID control to predict offset trends in existing laser beam pointing stabilization control methods.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] like Figure 1 As shown in the figure, this application proposes a laser beam pointing stabilization control method based on LSTM-PID, which includes the following steps: S1: Acquire the spot position signal and control voltage signal.

[0026] S2: Construct a multi-dimensional time-series dataset based on the spot position signal and the control voltage signal.

[0027] S3: Based on the multidimensional time series dataset, train the LSTM prediction model to obtain the trained LSTM prediction model.

[0028] S4: Obtain the historical multidimensional time series data captured by the sliding window, and based on the historical multidimensional time series data, use the trained LSTM prediction model to predict the spot offset at the next moment, and generate the corresponding feedforward control quantity; the historical multidimensional time series data includes the current actual position of the spot and the control voltage at the corresponding moment.

[0029] S5: Based on the deviation between the current actual position of the light spot and the target zero point position, calculate the corresponding real-time feedback control quantity through PID closed-loop control.

[0030] S6: The feedforward control quantity and the real-time feedback control quantity are weighted and fused to output the final control voltage signal; the final control voltage signal is used to control the beam deflection actuator to deflect at an angle to compensate for the beam pointing error and achieve laser beam pointing stability.

[0031] This application embodiment utilizes the acquired spot position signal and control voltage signal to construct a multi-dimensional time-series dataset to train the LSTM prediction model. A sliding window is used to acquire historical multi-dimensional time-series data containing the current actual spot position and the corresponding control voltage. The trained LSTM prediction model predicts the spot offset at the next moment and generates a feedforward control quantity, achieving prediction of the spot offset trend and alleviating the lag problem associated with relying solely on PID closed-loop feedback for adjustment. Simultaneously, PID closed-loop control is used to calculate the real-time feedback control quantity based on the deviation of the current actual spot position from the target zero point position, achieving real-time correction of the current actual pointing error. The method proposed in this application embodiment weightedly fuses the feedforward control quantity and the real-time feedback control quantity to output the final control voltage signal, driving the beam deflection actuator to deflect at an angle. The feedforward prediction compensation and closed-loop feedback correction work synergistically to compensate for the beam pointing error, ultimately achieving stable control of the laser beam pointing.

[0032] As an optional implementation, in step S1, the acquired spot position signal and control voltage signal include a four-dimensional time series consisting of the spot's X-axis position, Y-axis position, X-axis control voltage, and Y-axis control voltage. This multi-dimensional time series dataset, composed of the spot position signal and control voltage signal, is used to train the LSTM prediction model. After the LSTM prediction model is trained, it predicts the spot's position at the next time step, i.e., the spot's X-axis position and Y-axis position at the next time step.

[0033] As an optional implementation, in steps S2 and S4, the multidimensional time series dataset and the multidimensional time series data at each moment in the historical multidimensional time series data both include the X-axis position of the light spot, the Y-axis position of the light spot, the X-axis control voltage, and the Y-axis control voltage.

[0034] As an optional implementation, in steps S3 and S4, the LSTM prediction model takes a continuous multi-dimensional time sequence of a preset length (specifically a four-dimensional time sequence consisting of the X-axis position of the light spot, the Y-axis position of the light spot, the X-axis control voltage, and the Y-axis control voltage) as input, outputs the predicted position of the light spot at the next moment, and determines the light spot offset at the next moment based on the predicted position of the light spot at the next moment and the actual position of the current light spot.

[0035] As an optional implementation, in step S3, based on the multidimensional time-series dataset, the LSTM prediction model is trained to obtain a trained LSTM prediction model, specifically including the following steps: S31: Based on the multidimensional time series dataset, a continuous multidimensional time series sequence of a preset length is extracted using a time sliding window as the model input, and the spot position corresponding to the next moment is used as the label to construct a time series sample.

[0036] S32: Standardize the time series samples to obtain standardized time series samples.

[0037] S33: Divide the standardized time series samples into a training set and a test set according to the time series ratio.

[0038] S34: Based on the training set, using a network structure containing at least one LSTM layer and a Dropout regularization layer, the LSTM prediction model is obtained through multiple rounds of training using an optimization algorithm. The LSTM prediction model is then tested based on the test set. After training and testing are completed, the trained LSTM prediction model is obtained.

[0039] As an optional implementation, in step S5, based on the deviation between the current actual position of the light spot and the target zero point position, the corresponding real-time feedback control quantity is calculated through PID closed-loop control, specifically including the following steps: S51: Calculate the proportional term, integral term, and differential term respectively based on the deviation between the actual position of the current light spot and the target zero point position.

[0040] The target zero-point position is the center point of the position-sensitive detector, corresponding to the origin (0, 0) of the light spot position coordinate system, i.e., the zero-point position where both the X-axis and Y-axis positions are 0. The deviation between the current actual position of the light spot and the target zero-point position is the positional deviation of the current actual coordinates (X, Y) of the light spot relative to the origin (0, 0) of the coordinate system. Specifically, the X-axis deviation is equal to the X-axis position of the light spot minus 0, and the Y-axis deviation is equal to the Y-axis position of the light spot minus 0.

[0041] S52: Calculate the feedback control voltage value based on the proportional term, the integral term, and the derivative term.

[0042] S53: Determine the real-time feedback control quantity based on the feedback control voltage value.

[0043] As an optional implementation, in step S6, when the feedforward control quantity and the real-time feedback control quantity are weighted and fused, corresponding weight coefficients are set for the feedforward control quantity and the real-time feedback control quantity respectively. After weighted superposition, the final control voltage value is obtained, and the corresponding final control voltage signal is generated according to the final control voltage value.

[0044] As an optional implementation, the method further includes: pre-determining a deflection threshold of the beam deflection actuator based on the beam transmission distance and the maximum deflection angle of the beam deflection actuator; the deflection threshold is used to ensure that the adjustable range of the beam deflection actuator can cover the beam jitter range, so as to meet the closed-loop negative feedback condition.

[0045] Based on the same inventive concept, this application also provides an LSTM-PID-based laser beam pointing stabilization control system for implementing the aforementioned LSTM-PID-based laser beam pointing stabilization control method. The solution provided by this system is similar to the implementation described in the above method; therefore, the specific limitations of the LSTM-PID-based laser beam pointing stabilization control system embodiments provided below can be found in the limitations of the LSTM-PID-based laser beam pointing stabilization control method described above, and will not be repeated here.

[0046] In one exemplary embodiment, such as Figure 2 As shown, a laser beam pointing stabilization control system based on LSTM-PID is provided. The system includes an optical path unit, a detection unit, a control unit, and an execution unit.

[0047] The optical path unit is used to generate a laser beam and collimate, transmit, split, and adjust the power of the beam so that the power of the beam incident on the detection unit is within the linear operating range.

[0048] The detection unit is used to acquire the position signal of the far-field light spot in real time and convert the position signal into an electrical signal and output it to the control unit.

[0049] The control unit includes an LSTM timing prediction module and a PID closed-loop control module. The LSTM timing prediction module is used to predict the spot offset at the next moment based on historical multi-dimensional timing data and generate a corresponding feedforward control quantity. The PID closed-loop control module is used to calculate the corresponding real-time feedback control quantity based on the deviation between the current actual spot position and the target zero point position. The control unit is also used to perform weighted fusion of the feedforward control quantity and the real-time feedback control quantity, output the final control voltage signal, and send it to the execution unit.

[0050] The execution unit is a beam deflection actuator, which is used to generate an angle deflection according to the final control voltage signal to compensate for beam pointing error and achieve laser beam pointing stability.

[0051] In this embodiment, the optical path unit includes a laser 1, a focusing lens 2, a high-reflection mirror group (including a first high-reflection mirror 3, a second high-reflection mirror 4, a third high-reflection mirror 10, and a fourth high-reflection mirror 11), a half-wave plate 5, a beam splitter 6, and a power meter 7.

[0052] The beam output from the laser 1 is collimated by the focusing lens 2, and then undergoes optical path adjustment and power distribution through the high-reflection mirror group, the half-wave plate 5 and the beam splitter 6 to output a split beam; the power meter 7 is used to monitor the power of the split beam in real time.

[0053] In this embodiment of the application, the detection unit includes a position-sensitive detector (including a first position-sensitive detector 12 and a second position-sensitive detector 14) and a beam positioning controller 15.

[0054] The position-sensitive detector is set at the far-field measuring point to collect the X-axis position and Y-axis position signals of the light spot in real time and transmit them to the beam positioning controller 15.

[0055] The beam positioning controller 15 is used to convert the X-axis position and Y-axis position signals of the light spot into electrical signals and output them to the control unit.

[0056] In this embodiment of the application, the beam deflection actuator includes a piezoelectric controller 16 and a feedback-enabled lens frame.

[0057] The piezoelectric controller 16 is used to drive the feedback frame to generate an angular deflection according to the final control voltage signal output by the control unit, to compensate for the beam pointing error in real time, and to stabilize the light spot near the target position.

[0058] In this embodiment, the optical path unit uses a 532nm all-solid-state single-frequency continuous laser 1 as the light source. It is sequentially equipped with a focusing lens 2 with a focal length of 200mm, multiple 532nm high-reflection mirrors (including a first high-reflection mirror 3, a second high-reflection mirror 4, a third high-reflection mirror 10, and a fourth high-reflection mirror 11), a half-wave plate 5, a 532nm beam splitter 6, a power meter 7, and a beam collector 13. This enables the collimation, transmission, beam splitting, power monitoring, and collection of the beam. The power of the beam incident on the position-sensitive detector is adjusted to a suitable range to ensure that the power of the beam entering the detection unit is within the linear operating range of the detection, thus avoiding saturation distortion.

[0059] In this embodiment, the detection unit uses a PDP90A type position-sensitive detector, in conjunction with a KPA101 type beam positioning controller 15, to collect the X-axis position signal and Y-axis position signal of the far-field beam in real time, and convert the position information into an electrical signal and output it to the control unit to realize the monitoring of beam offset.

[0060] In this embodiment, the detection unit acquires two-dimensional position signals of the light spot in real time (including the X-axis position signal and the Y-axis position signal of the light spot), and the control unit synchronously records the dual-axis control voltage signals (including the X-axis control voltage and the Y-axis control voltage) output to the actuator, forming a continuous four-dimensional raw time-series data stream. The four-dimensional raw time-series data stream is truncated through a sliding window, and a preset number of historical continuous sampling moments are selected. The light spot position and control voltage corresponding to each moment are combined into a fixed-length multi-dimensional time-series segment (i.e., the historical multi-dimensional time-series data truncated by the sliding window), which is used as the input data of the LSTM prediction model.

[0061] In this embodiment, the control unit serves as the core of the system and includes an LSTM timing prediction module and a PID closed-loop control module.

[0062] The LSTM time-series prediction module takes a four-dimensional time sequence consisting of the X position, Y position, X-axis control voltage, and Y-axis control voltage of the spot at 50 historical moments as input. Through the structure of a two-layer LSTM network and a Dropout regularization layer, it learns the long-term dependence of the beam offset, outputs the predicted position of the spot at the next moment, and generates a lead prediction feedforward signal.

[0063] The PID closed-loop control module takes the deviation between the current actual position of the light spot and the target zero point as input, and calculates the feedback control quantity in real time through proportional, integral and derivative operations. This is used to quickly eliminate the current error, suppress oscillation and ensure the steady-state accuracy of the system.

[0064] The control unit weighted and fused the LSTM feedforward signal and the PID feedback signal to obtain the final control voltage signal, realizing a composite control that combines advance prediction and real-time correction.

[0065] In this embodiment, the execution unit uses a KPC101 piezoelectric controller 16 to drive a POLARIS-K1PZ dual-axis feedback-enabled lens frame (including a first feedback-enabled lens frame 8 and a second feedback-enabled lens frame 9). Based on the final control voltage signal output by the control unit, the feedback-enabled lens frame is driven to generate a micro-angle deflection, which compensates for the beam pointing error in real time, so that the light spot is always stable near the target zero point, thereby achieving the purpose of stabilizing the beam pointing.

[0066] In this embodiment, the optical platform is first built and the hardware is initialized. Figure 2 It can be seen that the optical path uses a 532nm all-solid-state single-frequency continuous laser 1 as the light source. After the beam is collimated by a focusing lens 2 with a focal length of 200mm, it is adjusted by a high-reflection mirror group (including a first high-reflection mirror 3, a second high-reflection mirror 4, a third high-reflection mirror 10, and a fourth high-reflection mirror 11), a half-wave plate 5, and a polarizing beam splitter 6 to control the optical power incident on the PDP90A model position-sensitive detector to be within 125μW, ensuring that the position-sensitive detector operates in the linear range and does not saturate. Figure 3 It can be seen that at a working wavelength of 532nm, the responsivity of the position-sensitive detector is 0.32 (unit is A / W, Ampere / W). According to the technical specifications of the PDP90A model position-sensitive detector, the maximum photocurrent that the position-sensitive detector can reach before reaching saturation is 40μA. The power of the position-sensitive detector is calculated according to formula (1).

[0067] (1).

[0068] in, The maximum receivable optical power of the position-sensitive detector, in units of ; The responsivity of the position-sensitive detector is represented by a value of 0.32 A / W in this embodiment.

[0069] Combination Figure 4 The effective detection area of ​​the PDP90A model position-sensitive detector is 7.2mm × 7.2mm. The edge detection accuracy of the photosensitive surface of the position-sensitive detector is relatively low. To ensure the overall control accuracy of the system, the spot point needs to be limited to the effective area of ​​the center of the position-sensitive detector, and the physical diameter of the incident spot needs to be controlled between 0.2 and 7mm. Let the spot radius be r, then the effective linear range of the position-sensitive detector in the horizontal and vertical directions is ±(3.6-r)mm.

[0070] This embodiment of the application uses two position-sensitive detectors, namely a first position-sensitive detector 12 and a second position-sensitive detector 14. The two position-sensitive detectors are arranged at a certain distance along the beam transmission optical path. A single position-sensitive detector can only obtain the two-dimensional position of the beam spot on the detection plane and cannot distinguish between the translational drift and the angular pointing drift of the beam. By collecting the X-axis position and Y-axis position of the beam spot at different transmission distances by the two position-sensitive detectors respectively, and combining the optical path distance between the two position-sensitive detectors, the translational offset and angular pointing offset of the beam can be calculated simultaneously. This achieves decoupled detection of beam position error and pointing error, which can provide feedback input for PID closed-loop control and provide more complete time-series characteristics for LSTM prediction models, improve the model's ability to predict beam offset trends, and achieve joint stability of beam position and beam pointing.

[0071] For the detection part, the position-sensitive detector of model PDP90A is placed at the far-field measuring point. The position-sensitive detector is controlled by model KPA101 beam positioning controller 15 to collect the X-axis and Y-axis position signals of the light spot in real time and upload them to computer 17 for further data storage and processing.

[0072] For the execution section, a KPC101 piezoelectric controller 16 drives a POLARIS-K1PZ feedback-enabled lens mount, converting the control voltage into a micro-angle deflection of the lens mount to achieve beam pointing compensation. It is important to note that before the system is put into formal operation, the measurement distance needs to be determined. The deflection threshold of the feedback-enabled eyeglass frame is pre-defined. According to the specification manual for the POLARIS-K1PZ model feedback-enabled eyeglass frame, the KPC101 piezoelectric controller 16 controls the maximum deflection angle of the feedback-enabled eyeglass frame in one dimension to be 375 μrad. Figure 5It can be seen that when the angle of the feedback frame is... When it rotates, according to geometric optics properties, the angular displacement of its reflected beam will increase to 2. In other words, in the system design, the maximum rotation angle of the frame in a single direction can be fed back. At this point, the maximum rotation angle of the reflected light path is (750 μrad ≈ 0.043°), therefore, the maximum angular displacement of the reflected beam can reach 750 μrad. From the tangent function, the following relationship can be obtained: (2).

[0073] in, This represents the spot offset, and the maximum spot offset. At that time, calculate the critical distance. At this distance, the feedback frame can cover the entire detection field of view. However, due to the actual spot size and random beam jitter, ideal conditions cannot be used. It is necessary to confirm whether the adjustable range of the feedback frame can compensate for beam jitter before a closed-loop negative feedback can be established. This application's embodiment sets... This meets the requirements for closed-loop control.

[0074] In this embodiment of the application, during the dataset construction and LSTM prediction model training process, under closed-loop operation, four-dimensional time-series data of the far-field light spot's X-position, Y-position, X-axis control voltage, and Y-axis control voltage are continuously collected and saved as a CSV file. Figure 6 As shown, a continuous sequence of length 50 is extracted from a time sliding window as the model input, and the X and Y positions of the far-field light spot corresponding to the next time step are used as labels to construct time-series samples. The time-series sample data is normalized to zero mean to eliminate dimensional differences, and then divided into training and test sets in an 8:2 time-series ratio.

[0075] The X-axis control voltage refers to the voltage signal output from the control unit to the beam deflection actuator, used to drive the feedback-enabled lens frame to generate a micro-angle deflection in the X-axis direction; the Y-axis control voltage refers to the voltage signal output from the control unit to the beam deflection actuator, used to drive the feedback-enabled lens frame to generate a micro-angle deflection in the Y-axis direction. Each control cycle outputs a set of X-axis and Y-axis control voltages to compensate for pointing errors in the X and Y directions of the light spot, respectively. When acquiring time-series data, the X-axis and Y-axis positions of the light spot at each moment, together with the output X-axis and Y-axis control voltages at that moment, form a four-dimensional time-series sample, used to construct a multi-dimensional time-series dataset and as input to the online LSTM prediction model.

[0076] like Figure 6As shown, the LSTM prediction model in this embodiment employs a two-layer LSTM network combined with a Dropout regularization layer. The first LSTM layer contains 64 neurons for extracting temporal features, and the second LSTM layer contains 32 neurons for feature fusion. Dropout layers with a 0.2 deactivation rate are applied after both layers to prevent overfitting. The Adam optimizer is used for 50 training rounds with a batch size of 32, ultimately yielding an LSTM prediction model suitable for real-time prediction.

[0077] This embodiment employs an LSTM-PID composite control algorithm. The computer 17 reads the actual position of the current light spot from the position-sensitive detector in real time and calculates the deviation between the current actual position of the light spot and the target zero-point position. The PID closed-loop control module calculates the proportional, integral, and derivative terms based on the deviation and outputs a feedback control voltage to quickly eliminate the current error and suppress oscillations. Simultaneously, the LSTM prediction model reads historical four-dimensional time-series data from the last 50 moments to predict the possible offset position of the light spot at the next moment and calculates the LSTM feedforward voltage accordingly to achieve advance compensation. The control unit weightedly superimposes the LSTM feedforward voltage and the PID feedback voltage to obtain the final control voltage, which is then output to the KPC101 piezoelectric controller 16 to drive the feedback-enabled lens frame deflection to correct the beam direction. The system executes the above process in a high-frequency cyclic manner to continuously stabilize the beam position.

[0078] Figure 7 This is a schematic diagram showing the location of the system operating freely for 3 hours without feedback. Figure 8 The diagram shows the position after 3 hours of operation with LSTM-PID feedback. It can be seen that the system in this embodiment was compared under the conditions of 3 hours of free operation without feedback and 3 hours of operation with LSTM-PID feedback. The results show that without LSTM-PID feedback, beam drift is significant, with the maximum X-axis offset approaching 0.3 mm and the maximum Y-axis offset approaching 0.2 mm. After adding the LSTM-PID composite control feedback of this embodiment, the beam spot is highly concentrated near zero, and the fluctuation range is significantly narrowed. The average X-axis error decreased from 0.1109 mm to 0.0048 mm, and the average Y-axis error decreased from 0.0400 mm to 0.0022 mm. The stability in both X and Y directions improved by more than 94%. There was no significant drift during long-term operation, demonstrating stable control, rapid response, and significant suppression of beam pointing jitter, thus significantly improving the far-field pointing stability of the laser beam.

[0079] In this embodiment, the timing data of the beam spot position and driving voltage are first acquired to train the LSTM prediction model. Then, the beam spot offset at the next moment is predicted by the LSTM timing prediction module, and a feedforward control quantity is output. Next, the PID closed-loop control module calculates and outputs a real-time feedback control quantity based on the deviation between the actual beam spot position and the target zero point. Finally, the feedforward control quantity and the real-time feedback control quantity are weighted and fused to obtain the final control voltage value, which is used to drive the feedback-enabled lens frame to deflect, compensate for beam pointing errors, and achieve laser beam pointing stability. This forms a complete closed loop of data acquisition, timing prediction, feedback control, and execution correction, structurally solving the problems of lag, drift, and limited accuracy in traditional feedback control.

[0080] This application's embodiments leverage LSTM's ability to effectively capture the temporal characteristics of beam offset and predict the offset trend in the next moment. It upgrades traditional hysteresis feedback to predictive feedforward, significantly reducing system response delay and demonstrating excellent suppression capabilities for both rapid jitter and slow drift. LSTM feedforward handles advance compensation, while PID feedback handles real-time correction. The two work together to improve dynamic response speed while ensuring steady-state control accuracy, enabling long-term continuous operation without drift or oscillation. The system does not rely on a precise physical model; instead, it learns disturbance patterns through data-driven learning, reducing the impact of various disturbances such as mechanical vibration and temperature drift, allowing for stable operation in complex environments.

[0081] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0082] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A laser beam pointing stabilization control method based on LSTM-PID, characterized in that, The method includes: Acquire light spot position signals and control voltage signals; Based on the spot position signal and the control voltage signal, a multi-dimensional time-series dataset is constructed; Based on the multidimensional time series dataset, the LSTM prediction model is trained to obtain the trained LSTM prediction model. The historical multidimensional time series data captured by the sliding window is obtained, and based on the historical multidimensional time series data, the trained LSTM prediction model is used to predict the spot offset at the next moment and generate the corresponding feedforward control quantity; the historical multidimensional time series data includes the current actual position of the spot and the control voltage at the corresponding moment; Based on the deviation between the current actual position of the light spot and the target zero point position, the corresponding real-time feedback control quantity is calculated through PID closed-loop control. The feedforward control quantity and the real-time feedback control quantity are weighted and fused to output a final control voltage signal; the final control voltage signal is used to control the beam deflection actuator to deflect at an angle to compensate for the beam pointing error and achieve laser beam pointing stability.

2. The laser beam pointing stabilization control method based on LSTM-PID according to claim 1, characterized in that, The multidimensional time series dataset and the multidimensional time series data of each moment in the historical multidimensional time series data each include the X-axis position of the light spot, the Y-axis position of the light spot, the X-axis control voltage, and the Y-axis control voltage; The LSTM prediction model takes a continuous multidimensional time series of a preset length as input and outputs the predicted position of the light spot at the next moment.

3. The laser beam pointing stabilization control method based on LSTM-PID according to claim 1, characterized in that, Based on the aforementioned multidimensional time-series dataset, the LSTM prediction model is trained to obtain a trained LSTM prediction model, specifically including: Based on the multidimensional time series dataset, a continuous multidimensional time series sequence of a preset length is extracted using a time sliding window as the model input, and the spot position corresponding to the next moment is used as the label to construct a time series sample; The time series samples are standardized to obtain standardized time series samples; According to the time series ratio, the standardized time series samples are divided into training set and test set; Based on the training set, the LSTM prediction model is obtained by using a network structure containing at least one LSTM layer and a Dropout regularization layer and performing multiple rounds of training through an optimization algorithm. The LSTM prediction model is then tested based on the test set. After training and testing are completed, the trained LSTM prediction model is obtained.

4. The laser beam pointing stabilization control method based on LSTM-PID according to claim 1, characterized in that, Based on the deviation between the current actual position of the light spot and the target zero point position, the corresponding real-time feedback control quantity is calculated through PID closed-loop control, specifically including: Based on the deviation between the current actual position of the light spot and the target zero point position, calculate the proportional term, integral term, and differential term respectively; The feedback control voltage value is calculated based on the proportional term, the integral term, and the derivative term. The real-time feedback control quantity is determined based on the feedback control voltage value.

5. The laser beam pointing stabilization control method based on LSTM-PID according to claim 1, characterized in that, When the feedforward control quantity and the real-time feedback control quantity are weighted and fused, corresponding weight coefficients are set for the feedforward control quantity and the real-time feedback control quantity respectively. After weighted superposition, the final control voltage value is obtained, and the corresponding final control voltage signal is generated according to the final control voltage value.

6. The laser beam pointing stabilization control method based on LSTM-PID according to claim 1, characterized in that, The method further includes: pre-determining a deflection threshold of the beam deflection actuator based on the beam transmission distance and the maximum deflection angle of the beam deflection actuator; the deflection threshold is used to ensure that the adjustable range of the beam deflection actuator can cover the beam jitter range, so as to meet the closed-loop negative feedback condition.

7. A laser beam pointing stabilization control system based on LSTM-PID, characterized in that, The system is used to implement the laser beam pointing stabilization control method based on LSTM-PID as described in any one of claims 1-6, and the system includes an optical path unit, a detection unit, a control unit, and an execution unit; The optical path unit is used to generate a laser beam and collimate, transmit, split, and adjust the power of the beam so that the power of the beam incident on the detection unit is within the linear operating range. The detection unit is used to acquire the position signal of the far-field light spot in real time, and convert the position signal into an electrical signal and output it to the control unit; The control unit includes an LSTM timing prediction module and a PID closed-loop control module. The LSTM timing prediction module is used to predict the spot offset at the next moment based on historical multi-dimensional timing data and generate a corresponding feedforward control quantity. The PID closed-loop control module is used to calculate the corresponding real-time feedback control quantity based on the deviation between the current actual spot position and the target zero point position. The control unit is also used to perform weighted fusion of the feedforward control quantity and the real-time feedback control quantity, output the final control voltage signal, and send it to the execution unit. The execution unit is a beam deflection actuator, which is used to generate an angle deflection according to the final control voltage signal to compensate for beam pointing error and achieve laser beam pointing stability.

8. The laser beam pointing stabilization control system based on LSTM-PID according to claim 7, characterized in that, The optical path unit includes a laser, a focusing lens, a high-reflectivity mirror group, a half-wave plate, a beam splitter, and a power meter; The laser beam output from the laser is collimated by the focusing lens, and then passes sequentially through the high-reflectivity mirror group, the half-wave plate, and the beam splitter for optical path adjustment and power distribution, outputting a split beam; the power meter is used to monitor the power of the split beam in real time.

9. The laser beam pointing stabilization control system based on LSTM-PID according to claim 7, characterized in that, The detection unit includes a position-sensitive detector and a beam positioning controller; The position-sensitive detector is set at the far-field measuring point to collect the X-axis position and Y-axis position signals of the light spot in real time and transmit them to the beam positioning controller. The beam positioning controller is used to convert the X-axis position and Y-axis position signals of the light spot into electrical signals and output them to the control unit.

10. The laser beam pointing stabilization control system based on LSTM-PID according to claim 7, characterized in that, The beam deflection actuator includes a piezoelectric controller and a feedback-enabled lens frame; The piezoelectric controller is used to drive the feedback frame to generate an angular deflection according to the final control voltage signal output by the control unit, so as to compensate for the beam pointing error in real time and stabilize the light spot near the target position.