Fast steering mirror adaptive control method based on model prediction

Abstract

The invention discloses a model prediction-based fast steering mirror adaptive control method, and relates to the technical field of fast steering mirror control, and the method comprises the steps: building an adaptive control model based on the parameter information of a fast steering mirror, splitting the adaptive control model into a plurality of sub-control regions, and executing model simulation on each sub-control region in a preset simulation period; predicting a fast reflector offset state of each sub-control area based on a simulation result, taking the fast reflector offset state as a training parameter, and executing local optimal control in a period on each sub-control area by an adaptive control model; and packaging all working procedures of local optimal control into a control assembly and storing the control assembly into a shared data pool for adaptive control calling of the fast steering mirror in other periods. Through partition control, state prediction and cross-cycle component multiplexing, the control precision and response efficiency of the fast reflecting mirror are improved, the cost of computing and compiling resources is reduced, and the method is suitable for a fast reflecting mirror control scene with a high-precision requirement.

Description

Technical Field

[0001] This invention relates to the field of fast-reflecting mirror control technology, and more specifically to a fast-reflecting mirror adaptive control method based on model prediction. Background Technology

[0002] As a core component in high-precision optical systems, fast-reflecting mirrors are widely used in fields such as astronomical observation, laser communication, and adaptive optics. Their control precision and response speed directly determine the performance ceiling of the entire optical system.

[0003] Existing fast-reflecting mirror control methods mostly adopt a globally unified control strategy, that is, a single control model is constructed based on the overall parameters of the fast-reflecting mirror. However, the physical structure of the fast-reflecting mirror has spatial heterogeneity, and the frequency domain characteristics and dynamic response of different regions are different, making it difficult for a single control model to adapt to the personalized control needs of each region.

[0004] Meanwhile, traditional control methods do not establish a mechanism for reusing cross-cycle control experience. Each control cycle requires retraining the model and adjusting the control parameters, which not only increases control delay but also wastes computational resources. Due to the lack of prediction of the offset state of each region of the fast-reflecting mirror, the formulation of control strategies often relies on empirical parameters, making it difficult to achieve local optimal control for different regions. This results in insufficient overall control accuracy of the fast-reflecting mirror, which cannot meet the requirements of high-precision optical systems for mirror stability. Summary of the Invention

[0005] The purpose of this invention is to provide a model-predictive-based adaptive control method for fast-reflecting mirrors to address the shortcomings in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a fast-reflecting mirror adaptive control method based on model prediction, comprising the following steps: Step S1: Construct the corresponding adaptive control model based on the parameter information of the fast-reflection mirror. The adaptive control model is divided into several sub-control regions, and model simulation is performed on each sub-control region within a preset simulation cycle. Step S2: Based on the simulation results, predict the fast mirror offset state under different sub-control regions, and use the fast mirror offset state as the training parameters of the adaptive control model. The adaptive control model then performs local optimal control for each sub-control region within one cycle. Step S3: Encapsulate all the procedures of local optimal control within a cycle into a control component, and store the control component in a shared data pool. The fast-reflection adaptive control in other cycles selects the control component in the shared data pool for invocation.

[0007] In a preferred embodiment, the process of constructing the corresponding adaptive control model based on the parameter information of the fast-reflecting mirror includes: The parameters of the fast-reflecting mirror include its structural parameters, frequency response parameters, environmental condition parameters, and calibration feedback information. The parameters are processed using engineering modeling methods to construct an instance model of the fast-reflecting mirror. The instance model is then processed using digital twin technology to obtain a twin model of the fast-reflecting mirror. Several control points are selected on the twin model, and frequency sweep test, step test and ray tracing test are performed on the fast mirror at each control point in sequence to obtain the spatial feature vector and dynamic response parameters of the fast mirror at each control point. The twin model of the fast-reflecting mirror is constructed into an adaptive control model based on several control points.

[0008] In a preferred embodiment, the process of constructing an adaptive control model based on control points and then decomposing the adaptive control model into several sub-control regions includes: Spatial feature vectors are used to characterize the unique spatial coordinates of each control point on the fast-reflection mirror; Dynamic response parameters are used to characterize the operating conditions of the fast-reflection mirror at each control point. The spatial feature vectors of several control points are clustered based on their respective dynamic response parameters to obtain several cluster sets. For the spatial feature vectors of the dynamic response parameters of all control points in each cluster set, the virtual point corresponding to the mean vector of all spatial feature vectors is set as the logical cluster center of the corresponding cluster set. Map all spatial feature vectors to the twin model, transform the twin model into an adaptive control model, and connect all control points in each cluster except the cluster center to their respective cluster centers in sequence to construct the sub-control region of the adaptive control model on the fast-reflection mirror.

[0009] In a preferred embodiment, the process of performing model simulation for each sub-control region within a preset simulation cycle includes: A simulation cycle is preset. Within the simulation cycle, the adaptive control model is started synchronously to simulate the model of several sub-control regions on the fast-reflecting mirror. The model simulation includes task configuration, environment configuration, simulation operation and simulation evaluation. The task configuration is used to create a simulation task for a sub-control area. The simulation task includes several instruction signals for controlling the fast-reflecting mirror to complete the correct operation, as well as establishing disturbance conditions that interfere with the fast-reflecting mirror. Environment configuration is used to set up the simulation environment for performing simulation tasks; The simulation operation involves the fast-reflecting mirror performing a simulation task in the simulation environment based on several command signals. The position of the disturbance condition inserted into the command signals is set, and the response state, time domain information and frequency domain information of the simulation task are recorded after each disturbance condition is added. Once the simulation is completed, a simulation evaluation will be performed.

[0010] In a preferred embodiment, the process of predicting the fast-reflection mirror offset state under different sub-control regions based on model simulation results includes: A standard model of a fast-reflecting mirror is constructed, which is divided into several targets to be analyzed. Model parameters are set for each target to be analyzed, and an instability model of the entire fast-reflecting mirror composed of all targets to be analyzed is constructed based on the model parameters. The mirror offset of each state analysis point of the fast-reflecting mirror is determined based on the instability model and the standard model, and the mirror offset of several state analysis points is integrated as the fast-reflecting mirror offset state under the corresponding sub-control area.

[0011] In a preferred embodiment, the process of using the fast-reflection mirror offset state as training parameters for the adaptive control model, and having the adaptive control model perform local optimal control for each sub-control region within one cycle, includes: Integrate the fast-reflection mirror offset states of all sub-control regions as the training parameter set; The frequency of control requests in each sub-control area within a preset time window is counted. Based on the frequency of control requests, frequently used areas and ordinary areas are determined. A container queue is created for frequently used areas, and a data warehouse is created for ordinary areas. Connect the resource operation node and the container queue, construct the first communication channel, process all the instruction signals of the adaptive control model into a control script, and transmit the control script, along with the parsed resources, to the container queue through the first communication channel at the resource operation node. The operation is performed by the container to achieve local optimal control of the commonly used area in one cycle; Connect the resource operation node and the data warehouse to build a second communication channel. At the resource operation node, carry the parsed resources and control scripts and transmit them to the data warehouse through the second communication channel. Based on the parsed resources, process the control scripts to complete the local optimal control of the ordinary area in one cycle.

[0012] In a preferred embodiment, the process by which the container queue performs container execution operations on control scripts in frequently used areas based on parsed resources includes: The container queue consists of a number of container nodes. Each container node is used to process a sub-control area of ​​the frequently used area. The first container node in the container queue is used as the source node, and the remaining container nodes are used as additional nodes. Based on the parsing resources, the control script is compiled at the source node, and the compilation result is transmitted to the attached node. The attached node selects the appropriate part of the compilation content according to its own control requirements for the fast mirror, and creates the complement of the appropriate compilation content and its own control requirements as an auxiliary control script, which is used to execute the respective control operations of all container nodes on the fast mirror. Each container node integrates the adapted compiled content with the auxiliary control script to create its own historical call template, and caches it in its own container node. When the corresponding sub-control area performs local optimal control in the future, it will take priority to adapt from the historical call template.

[0013] In a preferred embodiment, the process of encapsulating all the steps of local optimal control within a cycle into a control component includes: Local optimal control is based on several processes that perform mirror control operations. The process duration of each process and the connection conditions before and after the process are obtained. The connection conditions before and after the process include parallel execution conditions and sequential execution conditions. Concurrent execution is enabled for several processes under parallel execution conditions, and the maximum operation timing axis and the remaining operation timing axis are set. Candidate processes are filtered based on the process duration of each process under sequential execution conditions. It is determined whether the candidate process meets the requirement of encapsulating the process under parallel execution conditions. If it meets the requirement, a corresponding sub-control component is generated. If it does not meet the requirement, the corresponding process is directly encapsulated and generated as a sub-control component. Based on the condition of minimizing the cumulative value of process duration, the sub-control components of all processes in the local optimal control within a cycle are integrated to obtain the control components corresponding to the sub-control region of the local optimal control.

[0014] In a preferred embodiment, the process of storing control components in a shared data pool and selecting control components from the shared data pool for invocation by fast-reflection mirror adaptive control in other cycles includes: Configure the database, set up several data interfaces for communication connection to the database, set up the management port to establish communication paths from each data interface to the management port, create a data sharing window at the management port, and treat the database under the data sharing window and the several data interfaces connecting to the database as a shared data pool. The control components are stored in the shared data pool, and the component information of each control component is used as the retrieval information credential. When the operation of the fast-reflection mirror adaptive control in other cycles needs to be executed, a data interface is selected as the information upload interface, and the operation information of the fast-reflection mirror adaptive control in other cycles is input into the shared data pool. Based on the matching retrieval information credentials in the shared data pool, the control component that conforms to the corresponding execution process of the fast-reflection mirror adaptive control under the current cycle is invoked.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs an adaptive control model based on the parameter information of the fast-reflecting mirror and divides it into several sub-control regions, thereby realizing the zonal control of the fast-reflecting mirror. It adapts to the structural characteristics and dynamic response differences of different regions, effectively improving the control specificity of each region and solving the problem of insufficient adaptability of traditional global unified control models. It uses the simulation results of the model to predict the offset state of the fast-reflecting mirror and uses it as training parameters to execute local optimal control. This makes the formulation of control strategies based on accurate state prediction data rather than empirical parameters, significantly improving the control accuracy of each sub-control region and ensuring the stability of the mirror.

[0016] 2. This invention encapsulates the locally optimal control procedures within a cycle into control components and stores them in a shared data pool, thereby enabling the reuse of cross-cycle control experience. Subsequent cycles can directly call the adapted control components without repeating model training and parameter debugging, reducing control latency and lowering computational resource consumption. Attached Figure Description

[0017] 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 recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, please refer to Figure 1 As shown in this embodiment, the fast-reflecting mirror adaptive control method based on model prediction includes the following steps: Step S1: Construct the corresponding adaptive control model based on the parameter information of the fast-reflection mirror. The adaptive control model is divided into several sub-control regions, and model simulation is performed on each sub-control region within a preset simulation cycle. Step S2: Based on the simulation results, predict the fast mirror offset state under different sub-control regions, and use the fast mirror offset state as the training parameters of the adaptive control model. The adaptive control model then performs local optimal control for each sub-control region within one cycle. Step S3: Encapsulate all the procedures of local optimal control within a cycle into a control component, and store the control component in a shared data pool. The fast-reflection adaptive control in other cycles selects the control component in the shared data pool for invocation.

[0021] It should be further explained that, in the specific implementation process, the process of constructing a corresponding adaptive control model based on the parameter information of the fast-reflection mirror, and decomposing the adaptive control model into several sub-control regions, includes: The parameters of the fast-reflecting mirror include its structural parameters, frequency response parameters, environmental condition parameters, and calibration feedback information. The structural parameters of the fast-reflecting mirror are measured at the factory. These include the moment of inertia of the mirror and rotating components about the axis of rotation, the stiffness coefficient K_s and damping coefficient C_s of the flexible hinge, the current-torque conversion coefficient of the corresponding drive unit (voice coil motor), the voltage-displacement conversion coefficient of the position detector (PSD, encoder), and the effective aperture and maximum mechanical deflection angle of the mirror. The frequency response parameters describe the real-time dynamic behavior of the fast-reflecting mirror, specifically including the resonant frequency and damping ratio, equivalent stiffness and damping. The environmental state parameters describe the environment in which the fast-reflecting mirror is located and its real-time operating point, and are the external factors that cause changes in the model. The conditions include the environment and the temperature of the mirror and the actuator (the main factors causing the drift of mechanical parameters such as stiffness and resonant frequency), the operating point state, including the current deflection angle, angular velocity and current control quantity of the fast-reflecting mirror (represented by drive voltage or current); the calibration feedback information is used to verify and calibrate the adaptive control model corresponding to the fast-reflecting mirror, which includes the tracking error represented by the difference between the command position and the actual output position, the saturation state of the control quantity indicating whether the actuator is working in the nonlinear saturation region, and the dynamic response parameters characterized by the actual measured step response rise time, overshoot and settling time; The parameter information of the fast-reflecting mirror is processed by engineering modeling methods to construct an instance model corresponding to the fast-reflecting mirror. The instance model is then processed by digital twin technology to obtain a twin model corresponding to the fast-reflecting mirror. Specifically, a real-time data synchronization channel is established between the virtual twin model and the fast-reflecting mirror in the actual scenario. This channel is used for real-time data interaction between the twin model and the fast-reflecting mirror. Specifically, based on a preset interaction frequency or triggering event, the real-time environmental state parameters (such as ambient temperature and real-time deflection angle), calibration feedback information (such as tracking error), and dynamic response parameters of the fast-reflecting mirror in the actual scenario are updated to the twin model in real time. Based on the real-time updated twin model, the corresponding parameters of the adaptive control model are calibrated online, ensuring real-time data synchronization between the virtual twin model and the fast-reflecting mirror in the actual scenario.

[0022] Several control points are selected on the twin model corresponding to the fast-reflecting mirror. At each control point, the fast-reflecting mirror is subjected to frequency sweep test, step test and ray tracing test in sequence to obtain the spatial feature vector and dynamic response parameters of the fast-reflecting mirror at each control point. To further explain, within the mirror area of ​​the fast-reflecting mirror corresponding to the twin model, control points are selected using a uniform grid method (e.g., setting one point at the center of every 10mm x 10mm grid) or a key feature point method (e.g., directly above the flexible hinge, at the center of the mirror, or at the edge). The total number of control points is between N1 and N2 (e.g., for a fast-reflecting mirror with a diameter of 100mm, N1=20, N2=100) to ensure sufficient sampling of the dynamic response characteristics of the mirror.

[0023] Spatial feature vectors are used to characterize the unique spatial coordinates of each control point on the fast-reflecting mirror; Dynamic response parameters are used to characterize the operating conditions of the fast-reflecting mirror at each control point (such as different resonant frequencies, lens angular displacement, angular velocity, or moment of inertia). The spatial feature vectors corresponding to several control points are clustered based on their respective dynamic response parameters to obtain several cluster sets. Each cluster set includes several spatial feature vectors with the same working condition characteristics. The logical clustering center is divided based on the dynamic response parameters of different spatial feature vectors. Different spatial feature vectors are divided into clusters based on the similarity of dynamic response parameters. The mean vector (centroid of different spatial feature vectors) of all spatial feature vectors included in each cluster is obtained by the K-means algorithm. It can be seen that the location of the cluster center is not the same as the physical center of the fast-reflecting mirror. Map the spatial feature vectors of all clusters to the Siamese model, convert the Siamese model into an adaptive control model, and connect the control points in each cluster except the cluster center to their respective cluster centers in sequence, thereby constructing different sub-control regions on the fast-reflection mirror corresponding to the adaptive control model; The system includes a partition reassessment cycle. When the duration of the previous sub-control area division reaches one partition reassessment cycle, the dynamic response parameters of each control point of the fast-reflecting mirror are re-collected, and cluster analysis is performed again. If the similarity between the newly generated sub-control area division and the current division is lower than a preset threshold, the new partitioning results are used for updating, thereby changing the different sub-control areas on the fast-reflecting mirror. The purpose is to solve the problem that the dynamic response parameters of the fast-reflecting mirror will change with the environment (temperature) and operating conditions (long-term wear and tear). Fixed partitioning will lead to a mismatch between the later partitioning and the actual dynamic response characteristics, resulting in a gradual decrease in control accuracy and failure to meet the adaptive requirements.

[0024] It should be noted that for the spatial feature vector of the dynamic response parameters of all control points within each cluster, the virtual point corresponding to the mean vector of all spatial feature vectors is set as the logical cluster center of the corresponding cluster. The current logical cluster center is mapped to the nearest actual control point on the twin model of the fast-reflection mirror, which serves as the proxy control point for the sub-control region. The proxy control point is used for subsequent regional connection and simulation scheduling of the corresponding sub-control region.

[0025] It should be further explained that, in the specific implementation process, the process of performing model simulation for each sub-control region within a preset simulation cycle includes: A simulation cycle is preset, and the simulation cycle is denoted as . =[ , ];in, This is the start time of a preset simulation cycle. This is the end time of the simulation cycle; During the simulation cycle, the adaptive control model is synchronously started to simulate the model of several sub-control regions on the fast-reflecting mirror. The model simulation includes task configuration, environment configuration, simulation operation and simulation evaluation. The task configuration is used to create a simulation task for each sub-control area. The simulation task includes several instruction signals for controlling the fast-reflecting mirror to complete the correct operation, as well as establishing disturbance conditions to interfere with the fast-reflecting mirror. The environment configuration is used to build a simulation environment for executing simulation tasks and to monitor the simulation environment in real time. When the simulation environment does not meet the preset safety standards, the current simulation environment is reconfigured. The simulation operation involves the fast-reflecting mirror performing a simulation task in the simulation environment based on several command signals. The position of the disturbance condition inserted into the command signals is set, and the response state, time domain information and frequency domain information of the simulation task are recorded after each disturbance condition is added. During simulation, a coupling transfer function is introduced between different sub-control regions. The coupling transfer function is used to simulate the command signal or disturbance condition of a sub-control region and the combined effect of the two on the adjacent sub-control regions. The dynamic response data of each sub-control region after the influence is recorded. If the response status is no response, no action will be taken; If the response status is a successful response, the time domain information and frequency domain information are used as the performance indicators corresponding to the current closed-loop simulation moment. The obtained performance indicators are used as the reference indicators corresponding to the next instruction signal. The simulation operation corresponding to the next instruction signal is completed, and the indicator change trends corresponding to different instruction signals throughout the entire simulation task are compared as the operation information of the simulation operation corresponding to each sub-control area. After the simulation job is completed, the job information of each sub-control area is set as its own simulation final result. The simulation final result within the current preset simulation cycle is evaluated. If the simulation final result of a certain sub-control area does not meet the preset training stop criteria, the simulation task and simulation environment are reconfigured until the simulation final result meets the training stop criteria.

[0026] It should be noted that the training stoppage criteria include: the step response overshoot of the simulation output is less than 5%, the settling time is less than a preset millisecond value (e.g., 1 millisecond), and the frequency response curve within the target frequency band (e.g., >800Hz) has a good fit with the ideal curve of more than 90%. When the final simulation result meets all the preset conditions simultaneously, it is determined to meet the training stoppage criteria.

[0027] It should be further explained that, in the specific implementation process, the fast-reflecting mirror offset state is predicted based on the model simulation results under different sub-control regions. The fast-reflecting mirror offset state is then used as the training parameters of the adaptive control model. The process by which the adaptive control model performs local optimal control for each sub-control region within one cycle includes: A standard model corresponding to the fast-reflecting mirror under normal working conditions is constructed. Based on the different sub-control regions where the fast-reflecting mirror is located, the fast-reflecting mirror is divided into several targets to be analyzed. The simulation results of the fast-reflecting mirror corresponding to the model in different sub-control regions are used as the model parameters of each target to be analyzed. Then, based on the model parameters, the instability model of the entire fast-reflecting mirror composed of all the targets to be analyzed is constructed. The instability model is used to characterize the real-time mirror position of the fast-reflecting mirror in different sub-control areas; the standard model is used to characterize the standard mirror position of each sub-control area when the fast-reflecting mirror is in a standard scene. In constructing the instability model, the disturbance conditions recorded in the simulation operation and the dynamic response data of the disturbance conditions to each sub-control region are used as model parameters. The constructed parameters obtained after integration are used to characterize the comprehensive offset characteristics under the combined effect of structural differences and disturbance condition responses. This solves the problem of coupling interference between different sub-control regions (e.g., the deflection of sub-control region A may cause a small deformation of sub-control region B). The model parameters include the disturbance conditions of different sub-control regions, which can avoid the problem of large deviation between simulation results and actual working conditions caused by coupling interference, and ensure the accuracy of offset state prediction and local optimal control of the fast-reflecting mirror. The real-time mirror position and the standard mirror position of the unstable model and the standard model at the same coordinate position are merged into a single state analysis point. The mirror offset corresponding to each state analysis point is determined. The mirror offsets of a number of state analysis points of the fast-reflecting mirror in each sub-control area are integrated as the fast-reflecting mirror offset state in the corresponding sub-control area.

[0028] The fast-reflecting mirror offset state is used to characterize the degree of offset of the fast-reflecting mirror relative to the standard position in the corresponding sub-control area. The fast-reflecting mirror offset states of all sub-control areas are integrated as a training parameter set, and the area label corresponding to each sub-control area is marked in the training parameter set. The area label of several sub-control areas is denoted as i, then i = 1, 2, 3, ..., n, where n is a natural number greater than 0. Based on the historical simulation and control data corresponding to the fast-reflection mirror, the control request frequency of each sub-control area within the preset time window is counted. Sub-control areas with frequencies exceeding the preset threshold are marked as frequently used areas, and sub-control areas with other frequencies are marked as ordinary areas. Create container queues for frequently used regions and data warehouses for ordinary regions; Connect the resource operation node and the container queue, construct the first communication channel, process all the instruction signals of the adaptive control model into a control script, and transmit the control script, along with the parsed resources, to the container queue through the first communication channel at the resource operation node. The data warehouse destroys the corresponding compiled data after processing the control script based on the parsing resources, thereby releasing the storage space corresponding to the data warehouse. The container queue, on the other hand, does not destroy the compiled data. It sets a capacity limit for the container queue cache, such as 80% of the maximum storage capacity. The container node adopts a cache management strategy based on the Least Recently Used (LRU) algorithm or call frequency weighting for historical call templates. When the cache space reaches the 80% limit, the least frequently used template is automatically eliminated. At the same time, a lifespan timer is set for each template. Templates that have not been used for more than the preset time limit will also be automatically cleaned up.

[0029] The container queue completes the container execution operation of the control script in the common area based on the parsing resources. The specific content of the container execution operation is as follows: the container queue consists of a number of container nodes. Each container node is used to process any sub-control area belonging to the common area (in the common area, the fast mirror frequently executes control operations). The first container node of the container queue is used as the source node, and the remaining container nodes are used as additional nodes to execute message broadcasting. When the control script is transmitted to the container queue through the first communication channel, the control script is compiled at the source node based on the parsing resources, and the compilation result is transmitted to several additional nodes corresponding to the source node. The additional nodes select the appropriate part of the compiled content based on their own control requirements for the fast mirror, and create an auxiliary control script with the complement of the adapted compiled content and their own control requirements. This script is then used to execute the partial control operations of the fast mirror for all container nodes to complete the local optimal control of each sub-control area. Before the auxiliary control script is generated or executed by the additional node, the auxiliary control script is verified. Based on the physical constraint rule base of the fast mirror (such as maximum drive current, mechanical motion interference envelope) and timing logic rules, it is detected whether there are drive instruction conflicts or resource competitions between different auxiliary control scripts. If a conflict is detected, the control instructions of some auxiliary control scripts are replanned according to the preset priority or through optimization algorithm to eliminate the conflict. Each container node in the container queue integrates its own adapted compiled content with the auxiliary control script to create its own historical call template, which is then cached in its own container node. When the corresponding sub-control area performs local optimal control, it will prioritize adaptation from the historical call template.

[0030] The operation is performed by the container to achieve local optimal control of the commonly used area in one cycle; Connect the resource operation node and the data warehouse to build a second communication channel. Similarly, the resource operation node carries the parsed resources and control scripts and transmits them to the data warehouse through the second communication channel. Based on the parsed resources, the control scripts are processed to complete the local optimal control of the ordinary area in one cycle.

[0031] It should be noted that the adaptive control model is trained offline using training parameters. Specifically, the training samples are: the offset state of the fast-reflecting mirror in the sub-control area is used as training samples, supplemented by historical simulation data of the fast-reflecting mirror (sample size ≥ 10,000 sets) to construct a complete offline training dataset, actual historical control data (including control script execution records and conflict handling results), and simulation data of preset disturbance scenarios (covering the full range of disturbance intensity from 0.01g to 0.1g and frequency from 10Hz to 100Hz). The training samples must cover fully coupled disturbance scenarios such as simultaneous offset of several sub-control areas and single-area deflection causing deformation of other areas to ensure the comprehensiveness of the samples.

[0032] The training samples are divided into a 7:2:1 ratio, namely a training set (for model training), a validation set (for parameter tuning), and a test set (for final validation) to avoid overfitting. Offline training is conducted in stages, including a model initialization training stage and a sub-control region-specific training stage. Model initialization training is performed based on the difference in mirror offset between the standard model and the unstable model, training the model to learn the correspondence between offset states and control commands. The training parameters are: initial learning rate of 0.001, 500 iterations, batch size of 64, and the Adam optimizer. The sub-control region-specific training is used to train common and ordinary regions separately to enhance model adaptability. The specific execution process is as follows: the training content for common regions includes: training... The system incorporates capabilities for high-frequency control scenario adaptation, rapid compilation of control scripts, historical template adaptation and invocation, and conflict detection and handling. Combined with a container queue caching management strategy, the model learns the template eviction logic of the LRU algorithm. The corresponding training parameters are set as follows: learning rate 0.0005, iterations 300, batch size 32, and a target response latency of ≤3ms as an auxiliary training metric. The training for general areas focuses on adapting to low-frequency control scenarios, data warehouse resource release logic, and efficient control script processing capabilities to ensure control stability. Training parameters are consistent with those for commonly used areas, and a target utilization rate of ≥85% is set as an auxiliary training metric. The adaptive control model, after offline training, is subsequently encapsulated into corresponding control components for online invocation during fast-reaction mirror control execution in other cycles.

[0033] It should be further explained that, in the specific implementation process, the process of encapsulating all the procedures for local optimal control within a cycle into a control component includes: The local optimal control is based on several processes that perform mirror control operations. The process duration and the connection conditions before and after the process are obtained for each process. The connection conditions before and after the process include parallel execution conditions and sequential execution conditions. It should be noted that processes under the condition of parallel execution can be executed in parallel without any restrictions on the order of execution. However, processes under the condition of sequential execution are subject to logical restrictions on execution. The sequential execution condition is used to restrict the order of execution of different processes. For example, process Step 1 must be executed first, process Step 3 must be executed after process Step 2, and process Step 4 must be executed before process Step 4.

[0034] Concurrent execution is enabled for several processes under parallel execution conditions. The process with the shortest execution time among the concurrently executed processes is taken as the window start point, and the process with the longest execution time is taken as the window end point. The maximum operation timing axis is constructed based on the window start point and the window end point. In the concurrently executed processes, apart from the processes with the longest and shortest durations, the remaining processes are compared sequentially with the longest operation timeline to obtain the remaining operation timeline for each process. Obtain the process duration for each process under the sequential execution condition, and determine whether the process duration is less than or equal to the maximum operation time axis; If so, the qualified operation is stored as a candidate operation on the maximum operation timing axis, and it is determined whether there are any remaining operation timing axes that are suitable for the execution of the current candidate operation. If yes, then select the appropriate remaining operation time axis to store the current candidate process, and perform one encapsulation of the current candidate process and the processes under concurrent execution conditions at adjacent positions to construct a corresponding sub-control component. If no, then select the next candidate process and repeat the above operation until all candidate processes that adapt to the remaining operation time axis have been processed. Among them, candidate processes that fail to adapt to the remaining operation time axis are encapsulated on the maximum operation time axis to generate corresponding sub-control components. Integrate all sub-control components corresponding to the processes of the local optimal control within a cycle. The integration is based on the minimum cumulative value of the process duration, thereby obtaining the control component corresponding to the sub-control region of the local optimal control. If not, the corresponding process will be directly encapsulated to generate the corresponding sub-control components.

[0035] It should be noted that all processes corresponding to the local optimal control within a cycle are encapsulated based on the minimum cumulative process duration, so as to minimize the total encapsulation time overhead. Secondly, processes are classified based on different process connection conditions, and the maximum operation timing axis and the remaining operation timing axis are set so that different processes are encapsulated with the minimum time overhead and number of times, further reducing the overhead caused by encapsulation and improving the efficiency of component encapsulation.

[0036] It should be further explained that, in the specific implementation process, the process of storing the control components in the shared data pool and then selecting the control components from the shared data pool for invocation by the fast-reflection mirror adaptive control in other cycles includes: Configure the database, synchronously configure several data interfaces corresponding to the database, bind corresponding data sharing permissions to each data interface, set up a management port, establish communication paths from each data interface to the management port, and establish communication connections between each data interface and the database. Create a data sharing window at the management port, and treat the database under the data sharing window and several data interfaces connecting to the database as a shared data pool; A corresponding situation monitoring envelope network is established for the shared data pool. The situation monitoring envelope network consists of several monitoring envelope regions. Each monitoring envelope region is used to determine whether there is an abnormal situation when receiving control components. Each monitoring envelope region transmits its own monitoring logs to the management port in real time. The management port determines the situation of the monitoring envelope region based on the situation rules formulated based on historical data. If so, an envelope path is established between the current abnormal situation monitoring envelope area and other envelope areas in normal situation. The control components are transmitted to the envelope area in normal situation through the envelope path, and the control components are stored by the shared data pool. If not, the control component will receive the data directly from the current monitoring envelope area and store it in the shared data pool; It should be noted that the role of the envelope path in the situational awareness envelope network of the shared data pool is to establish a dedicated data transmission channel between the abnormal and normal areas when the target monitoring envelope area of ​​the receiving control component experiences an abnormal situation and cannot complete the reception and verification work normally. This is a key link to ensure that the control component can bypass the abnormal node and access the shared data pool securely and efficiently. It also plays an auxiliary role in transmission control, path tracing, and data verification. Specifically, its functions include: solving the transmission and processing limitations of the abnormal monitoring envelope area, enabling the control component to transmit directionally and without interference from the abnormal node to the normal node, ensuring that the control component will not experience upload failures or storage interruptions due to the situational anomalies of a single / local monitoring envelope area, and ensuring the continuity of the shared data pool's reception process for the control component; as a dedicated transmission path under abnormal situations, it is distinguished from the regular transmission link in the normal area, which can prevent the situational problems in the abnormal area (such as data transmission delays, verification logic failures, network fluctuations, etc.) from spreading to the normal envelope area. At the same time, it marks the control components transmitted within the path to ensure the integrity and uniqueness of the data during transmission and prevent the loss or tampering of component information. In addition, the envelope path is used to synchronously record cross-regional transmission information of control components (such as the abnormal region identifier of the transmission initiation, the normal region identifier of the reception, the transmission time, the component unique code, etc.), and synchronize this information to the management port. This provides data support for the management port to judge the impact range of abnormal situations and locate abnormal nodes. It also leaves traceable transmission records for subsequent source tracing of control components and condition review of shared data pools.

[0037] Each control component stored in the shared data pool uses its own component information as a retrieval information credential. When the operation of the fast-reflection mirror adaptive control in other cycles needs to be executed, it selects a data interface as the information upload interface and inputs the operation information of the fast-reflection mirror adaptive control in other cycles into the shared data pool. Based on the matching of operation information with the appropriate retrieval information credentials in the shared data pool, the control components that conform to the corresponding execution process of the fast-reflection mirror adaptive control under the current cycle are invoked.

[0038] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A fast-reflecting mirror adaptive control method based on model prediction, characterized in that, Includes the following steps: Step S1: Construct the corresponding adaptive control model based on the parameter information of the fast-reflection mirror. The adaptive control model is divided into several sub-control regions, and model simulation is performed on each sub-control region within a preset simulation cycle. Step S2: Based on the simulation results, predict the fast mirror offset state under different sub-control regions, and use the fast mirror offset state as the training parameters of the adaptive control model. The adaptive control model then performs local optimal control for each sub-control region within one cycle. Step S3: Encapsulate all the procedures of local optimal control within a cycle into a control component, and store the control component in a shared data pool. The fast-reflection adaptive control in other cycles selects the control component in the shared data pool for invocation.

2. The fast-reflecting mirror adaptive control method based on model prediction according to claim 1, characterized in that, The process of constructing the corresponding adaptive control model based on the parameter information of the fast-reflection mirror includes: The parameters of the fast-reflecting mirror include its structural parameters, frequency response parameters, environmental condition parameters, and calibration feedback information. The parameters are processed using engineering modeling methods to construct an instance model of the fast-reflecting mirror. The instance model is then processed using digital twin technology to obtain a twin model of the fast-reflecting mirror. Several control points are selected on the twin model, and frequency sweep test, step test and ray tracing test are performed on the fast mirror at each control point in sequence to obtain the spatial feature vector and dynamic response parameters of the fast mirror at each control point. The twin model of the fast-reflecting mirror is constructed into an adaptive control model based on several control points.

3. The fast-reflecting mirror adaptive control method based on model prediction according to claim 2, characterized in that, The process of constructing an adaptive control model based on control points and then decomposing the adaptive control model into several sub-control regions includes: Spatial feature vectors are used to characterize the unique spatial coordinates of each control point on the fast-reflection mirror; Dynamic response parameters are used to characterize the operating conditions of the fast-reflection mirror at each control point. The spatial feature vectors of several control points are clustered based on their respective dynamic response parameters to obtain several cluster sets. For the spatial feature vectors of the dynamic response parameters of all control points in each cluster set, the virtual point corresponding to the mean vector of all spatial feature vectors is set as the logical cluster center of the corresponding cluster set. Map all spatial feature vectors to the twin model, transform the twin model into an adaptive control model, and connect all control points in each cluster except the cluster center to their respective cluster centers in sequence to construct the sub-control region of the adaptive control model on the fast-reflection mirror.

4. The fast-reflecting mirror adaptive control method based on model prediction according to claim 3, characterized in that, The process of performing model simulation for each sub-control region within a preset simulation cycle includes: A simulation cycle is preset. Within the simulation cycle, the adaptive control model is started synchronously to simulate the model of several sub-control regions where the fast-reflecting mirror is located. The model simulation includes task configuration, environment configuration, simulation operation and simulation evaluation. The task configuration is used to create a simulation task for a sub-control area. The simulation task includes several instruction signals for controlling the fast-reflecting mirror to complete the correct operation, as well as establishing disturbance conditions that interfere with the fast-reflecting mirror. Environment configuration is used to set up the simulation environment for performing simulation tasks; The simulation operation involves the fast-reflecting mirror performing a simulation task in the simulation environment based on several command signals. The position of the disturbance condition inserted into the command signals is set, and the response state, time domain information and frequency domain information of the simulation task are recorded after each disturbance condition is added. Once the simulation is completed, a simulation evaluation will be performed.

5. The fast-reflecting mirror adaptive control method based on model prediction according to claim 4, characterized in that, The process of predicting the fast-reflection mirror offset state under different sub-control regions based on model simulation results includes: A standard model of a fast-reflecting mirror is constructed, which is divided into several targets to be analyzed. Model parameters are set for each target to be analyzed, and an instability model of the entire fast-reflecting mirror composed of all targets to be analyzed is constructed based on the model parameters. The mirror offset of each state analysis point of the fast-reflecting mirror is determined based on the instability model and the standard model, and the mirror offset of several state analysis points is integrated as the fast-reflecting mirror offset state under the corresponding sub-control area.

6. The fast-reflecting mirror adaptive control method based on model prediction according to claim 5, characterized in that, Using the fast-reflection mirror offset state as training parameters for the adaptive control model, the process of the adaptive control model performing local optimal control for each sub-control region within one cycle includes: Integrate the fast-reflection mirror offset states of all sub-control regions as the training parameter set; The frequency of control requests in each sub-control area within a preset time window is counted. Based on the frequency of control requests, frequently used areas and ordinary areas are determined. A container queue is created for frequently used areas, and a data warehouse is created for ordinary areas. Connect the resource operation node and the container queue, construct the first communication channel, process all the instruction signals of the adaptive control model into a control script, and transmit the control script, along with the parsed resources, to the container queue through the first communication channel at the resource operation node. The operation is performed by the container to achieve local optimal control of the commonly used area in one cycle; Connect the resource operation node and the data warehouse to build a second communication channel. At the resource operation node, carry the parsed resources and control scripts and transmit them to the data warehouse through the second communication channel. Based on the parsed resources, process the control scripts to complete the local optimal control of the ordinary area in one cycle.

7. The fast-reflecting mirror adaptive control method based on model prediction according to claim 6, characterized in that, The process by which the container queue performs operations on containers in frequently used areas based on parsed resources includes: The container queue consists of a number of container nodes. Each container node is used to process a sub-control area of ​​the frequently used area. The first container node in the container queue is used as the source node, and the remaining container nodes are used as additional nodes. Based on the parsing resources, the control script is compiled at the source node, and the compilation result is transmitted to the attached node. The attached node selects the appropriate part of the compilation content according to its own control requirements for the fast mirror, and creates the complement of the appropriate compilation content and its own control requirements as an auxiliary control script, which is used to execute the respective control operations of all container nodes on the fast mirror. Each container node integrates the adapted compiled content with the auxiliary control script to create its own historical call template, and caches it in its own container node. When the corresponding sub-control area performs local optimal control in the future, it will take priority to adapt from the historical call template.

8. The fast-reflecting mirror adaptive control method based on model prediction according to claim 7, characterized in that, The process of encapsulating all the steps of local optimal control within a cycle into a control component includes: Local optimal control consists of several processes that perform mirror control operations. The process duration of each process and the connection conditions before and after the process are obtained. The connection conditions before and after the process include parallel execution conditions and sequential execution conditions. Concurrent execution is enabled for several processes under parallel execution conditions, and the maximum operation timing axis and the remaining operation timing axis are set. Candidate processes are filtered based on the process duration of each process under sequential execution conditions. It is determined whether the candidate process meets the requirement of encapsulating the process under parallel execution conditions. If it meets the requirement, a corresponding sub-control component is generated. If it does not meet the requirement, the corresponding process is directly encapsulated and generated as a sub-control component. Based on the condition of minimizing the cumulative value of process duration, the sub-control components of all processes in the local optimal control within a cycle are integrated to obtain the control components corresponding to the sub-control region of the local optimal control.

9. The fast-reflecting mirror adaptive control method based on model prediction according to claim 8, characterized in that, The process of storing control components in a shared data pool and then selecting control components from the shared data pool for use by fast-reflection adaptive control in other cycles includes: Configure the database, set up several data interfaces for communication connection to the database, set up the management port to establish communication paths from each data interface to the management port, create a data sharing window at the management port, and treat the database under the data sharing window and the several data interfaces connecting to the database as a shared data pool. The control components are stored in the shared data pool, and the component information of each control component is used as the retrieval information credential. When the operation of the fast-reflection mirror adaptive control in other cycles needs to be executed, a data interface is selected as the information upload interface, and the operation information of the fast-reflection mirror adaptive control in other cycles is input into the shared data pool. Based on the matching retrieval information credentials in the shared data pool, the control component that corresponds to the execution process of the fast-reflection mirror adaptive control under the current cycle is invoked.

Images