Single module follow-up control method and system

Through multi-modal data processing and nonlinear system modeling, a multi-objective optimization control strategy is designed, which solves the problems of nonlinear characteristics and vibration suppression in single-module follow-up control, and achieves high accuracy, reliability and adaptability, improving Production efficiency and product quality.

CN118818992BActive Publication Date: 2024-11-22NANJING BISUNMIT INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411311354.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-22
Estimated Expiration
2044-09-20

AI Technical Summary

Technical Problem

The existing single-module follow-up control technology is difficult to effectively deal with the strong coupling nonlinear characteristics of the system, the vibration suppression effect is limited, it is difficult to adapt to complex and variable working conditions, the model prediction accuracy is insufficient, the state estimation is prone to deviation, and the adaptability is lacking. , it is difficult to deal with system parameter drift and environmental changes.

Method used

By acquiring multimodal data, extracting feature sets, building nonlinear system models, optimizing system states, designing multi-objective optimization control strategies, using adaptive solvers and parallel computing frameworks, generating robust control constraints, and real-time monitoring of system states is achieved. and adjust.

Benefits of technology

Achieve high-precision, high reliability and high-efficiency hopper motion control in complex industrial environments, improve production efficiency and product quality, reduce operating costs and safety risks, and enhance the system's adaptability and anti-interference ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118818992B_ABST
    Figure CN118818992B_ABST
Patent Text Reader

Abstract

The present invention discloses a single-module follow-up control method and system, comprising the following steps: multi-modal data acquisition and preprocessing, nonlinear system modeling and state estimation, predictive control strategy design, and adaptive optimization and control execution. The method processes multi-source sensor data through an adaptive algorithm, constructs an accurate nonlinear model using recursive least squares and Bayesian model selection, and implements robust state estimation using delayed coordinate embedding and adaptive fault-tolerant Kalman filtering. The predictive control strategy uses dynamic time domain adjustment and probabilistic constraint generation technology to improve control performance and robustness. Real-time optimization and robustness analysis are achieved through parallel computing and Monte Carlo simulation. The accuracy, stability and adaptability of the single-module follow-up control are improved, and it is suitable for high-speed and high-precision industrial automation applications.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention provides a follow-up control method, in particular a single-module follow-up control method and system. Background Art

[0002] The single-module follow-up control method has important research significance in the field of modern industrial automation. With the rapid development of intelligent manufacturing, the requirements for precision equipment control are constantly increasing, especially the demand for high-speed and high-precision motion control is becoming increasingly urgent. As a key technology, single-module follow-up control plays a vital role in the fields of counting machines, high-speed sorting machines, precision processing equipment, etc. It not only directly affects product quality and production efficiency, but is also closely related to energy efficiency, equipment life and safety.

[0003] At present, the research on single-module follow-up control mainly focuses on traditional PID control, fuzzy control and simple feedforward compensation. PID control is still widely used due to its simple and reliable characteristics. Researchers improve its performance through parameter self-tuning and multi-loop PID. Fuzzy control improves the system's adaptability to nonlinear and time-varying characteristics to a certain extent by introducing expert experience. Some researchers try to combine simple feedforward compensation with feedback control to improve the system's rapid response capability. In terms of vibration suppression, passive methods such as notch filters are mainly used to suppress vibrations of specific frequencies. For system modeling, most methods still rely on simplified linear models or lookup tables, which are difficult to accurately describe complex nonlinear dynamic characteristics. In terms of state estimation, conventional Kalman filters are used to process limited measurement noise, but their processing capabilities for strong nonlinear systems and non-Gaussian noise are limited.

[0004] Although these methods have achieved certain results in practice, there are still many technical problems in the face of increasingly stringent industrial needs. For example, it is difficult for traditional control methods to effectively handle the strongly coupled nonlinear characteristics of the system, especially when moving at high speeds, the control accuracy is significantly reduced due to the influence of nonlinear factors such as inertia and friction. In addition, the existing vibration suppression methods are mainly targeted at specific frequencies, and the suppression effect on wide-band and time-varying vibrations is limited, and it is difficult to adapt to complex and changing working conditions. Furthermore, simple system models cannot accurately describe the impact of factors such as material flow and temperature changes on system dynamics, resulting in insufficient model prediction accuracy. In addition, traditional state estimation methods are prone to estimation bias when facing sudden interference and sensor failures, affecting control performance. Another key issue is that existing methods are difficult to effectively handle multiple operating constraints (such as speed limits and safety boundaries) while meeting multiple control objectives (such as fast response, precise positioning, and minimization of energy consumption). Finally, most control strategies lack adaptive capabilities and are difficult to cope with long-term drift of system parameters and changes in the working environment, requiring frequent manual debugging and maintenance.

[0005] The existence of the above problems has seriously restricted the application and development of single-module follow-up control in the field of high-end manufacturing. Summary of the invention

[0006] The purpose of the invention is to provide a single-module follow-up control method and system to solve the above-mentioned problems existing in the prior art.

[0007] The technical solution is a single module follow-up control method, comprising the following steps:

[0008] Step S1, obtaining an original image sequence captured by a high-speed industrial camera, an original vibration signal collected by a vibration sensor, and hopper position data recorded by an encoder, and preprocessing them respectively to obtain a synchronized multimodal data set; extracting a feature set from the multimodal data set, and applying an adaptive parameter adjustment algorithm to initialize adjustable parameters according to the feature set and a predefined basic parameter set to obtain an initialized adjustable parameter set;

[0009] Step S2, based on the feature set and the adjustable parameter set, identify the system dynamic characteristics, obtain nonlinear system model parameters, build and optimize the nonlinear system model according to the nonlinear system model parameters, reconstruct the system state space based on the nonlinear system model, obtain the reconstructed state vector, combine the system model prediction and multi-source measurement data, estimate the system state, and obtain the estimated system state and covariance matrix;

[0010] Step S3, based on the estimated system state, the optimized nonlinear system model and the predefined control objectives, a model predictive control optimization problem is constructed; a dynamic time domain adjustment algorithm is applied to optimize the prediction and control time domains, and a probabilistic constraint generation algorithm is used to generate a robust control constraint set; a multi-objective optimization problem is constructed by comprehensively considering control performance, robustness and constraint satisfaction, and a complete optimization problem model is obtained;

[0011] Step S4: Based on the complete optimization problem model and the computing resource information currently available in the system, an adaptive solver selection algorithm is adopted and a parallel computing framework is used to solve the optimization problem, and the optimal control sequence is obtained and sent to each module; Monte Carlo simulation is performed to evaluate the robustness of the control strategy; control actions are executed to control the hopper movement, and the system response is monitored to obtain the actual system response, execution efficiency, and actual hopper control parameters.

[0012] According to another aspect of the present application, a single-module follow-up control system is also provided, comprising:

[0013] at least one processor; and,

[0014] a memory communicatively connected to at least one of the processors; wherein,

[0015] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the single-module follow-up control method described in any of the above technical solutions.

[0016] Beneficial effects, can achieve high-precision, high-reliability and high-efficiency operation in complex industrial environments. Not only improve production efficiency and product quality, but also reduce operating costs and safety risks. The successful application of this system will significantly improve the performance of precision equipment such as grain counters, and promote technological progress and industrial upgrading in related industries. Through single-module follow-up control, the grain counting machine can count grains more accurately. At the same time, the hopper and material conveyor belt move synchronously during the counting process, which improves the efficiency of the entire production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the present invention.

[0018] Figure 2 It is a flow chart of step S1 of the present invention.

[0019] Figure 3 It is a flow chart of step S2 of the present invention.

[0020] Figure 4 It is a flow chart of step S3 of the present invention.

[0021] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0022] like Figure 1 As shown, the single module follow-up control method is characterized by comprising the following steps:

[0023] Step S1, obtaining an original image sequence captured by a high-speed industrial camera, an original vibration signal collected by a vibration sensor, and hopper position data recorded by an encoder, and preprocessing them respectively to obtain a synchronized multimodal data set; extracting a feature set from the multimodal data set, and applying an adaptive parameter adjustment algorithm to initialize adjustable parameters according to the feature set and a predefined basic parameter set to obtain an initialized adjustable parameter set;

[0024] Step S2, based on the feature set and the adjustable parameter set, identify the system dynamic characteristics, obtain nonlinear system model parameters, build and optimize the nonlinear system model according to the nonlinear system model parameters, reconstruct the system state space based on the nonlinear system model, obtain the reconstructed state vector, combine the system model prediction and multi-source measurement data, estimate the system state, and obtain the estimated system state and covariance matrix;

[0025] Step S3, based on the estimated system state, the optimized nonlinear system model and the predefined control objectives, a model predictive control optimization problem is constructed; a dynamic time domain adjustment algorithm is applied to optimize the prediction and control time domains, and a probabilistic constraint generation algorithm is used to generate a robust control constraint set; a multi-objective optimization problem is constructed by comprehensively considering control performance, robustness and constraint satisfaction, and a complete optimization problem model is obtained;

[0026] Step S4: Based on the complete optimization problem model and the computing resource information currently available in the system, an adaptive solver selection algorithm is adopted and a parallel computing framework is used to solve the optimization problem, and the optimal control sequence is obtained and sent to each module; Monte Carlo simulation is performed to evaluate the robustness of the control strategy; control actions are executed to control the hopper movement, and the system response is monitored to obtain the actual system response, execution efficiency, and actual hopper control parameters.

[0027] In this embodiment, a comprehensive, efficient and robust single-module follow-up control system is realized through the organic combination of multimodal data acquisition and preprocessing, nonlinear system modeling and state estimation, predictive control strategy design, and adaptive optimization and control execution. The system can achieve high-precision hopper motion control and vibration suppression in a complex and changeable industrial environment. Specifically, multimodal data acquisition and preprocessing technology ensures that the system can obtain high-quality, synchronized visual, vibration and position data, providing a reliable data basis for subsequent modeling and control. Nonlinear system modeling and state estimation technology realizes accurate modeling and real-time state tracking of complex nonlinear systems through methods such as recursive least squares, Bayesian model selection and adaptive fault-tolerant Kalman filtering, providing accurate system models and state information for the design of control strategies. Predictive control strategy design technology realizes an optimization control strategy that considers multiple objectives and multiple constraints by constructing model predictive control problems, dynamically adjusting prediction and control time domains, and generating probabilistic constraints, which not only ensures control performance but also enhances the robustness of the system. Adaptive optimization and control execution technology achieves efficient solution and reliable execution of control strategies through parallel computing, robustness analysis and real-time monitoring feedback, ensuring that the system can achieve optimal control under real-time conditions.

[0028] Through accurate system modeling, real-time state estimation and optimization control strategy, the system can achieve high-precision trajectory tracking of the hopper motion. Even in high-speed motion or complex trajectory, it can maintain low tracking error to meet the requirements of precision control. Multimodal data fusion and predictive control strategy enable the system to effectively predict and suppress vibration during motion, significantly improving the smoothness and stability of the hopper motion. This not only improves positioning accuracy, but also reduces material loss and equipment wear caused by vibration. The system can automatically adjust model parameters, prediction time domain and control strategy according to changes in working conditions to adapt to different working environments and task requirements. This adaptability enables the system to maintain stable and efficient performance in long-term operation and reduces the need for manual intervention. By considering the uncertainty of state estimation, generating probabilistic constraints and performing robustness analysis, the system has strong anti-interference and fault tolerance capabilities. Even in the presence of model errors, parameter drift or external interference, it can maintain reliable control performance. Technologies such as parallel optimization calculation and dynamic time domain adjustment ensure that the system can complete complex optimization calculations in each control cycle and achieve true real-time optimization control. This enables the system to respond quickly to changes in working conditions and improves the timeliness and effectiveness of control. Robust constraint generation and real-time monitoring feedback mechanisms improve the safety of the system. Through constraint design that takes uncertainty into account and continuous performance monitoring, the system is able to avoid potential safety risks while ensuring high performance. Adaptive algorithms and parallel computing techniques optimize the use of computing resources, enabling the system to achieve high-performance control under limited hardware conditions. This not only reduces implementation costs, but also improves the energy efficiency of the system.

[0029] The high-performance single-module follow-up control in this embodiment can significantly improve the positioning accuracy, response speed and vibration suppression ability of the equipment, thereby improving product quality, reducing material waste and extending the service life of the equipment. It can also improve the flexibility and adaptability of the production system and provide technical support for the realization of intelligent and flexible production.

[0030] like Figure 2 As shown, the step S1 is specifically as follows:

[0031] Step S11, obtaining an original image sequence captured by a high-speed industrial camera; applying an adaptive exposure algorithm to the original image sequence, dynamically adjusting exposure parameters according to the overall brightness distribution of the image, until the contrast and detail visibility of the image reach a preset value, and obtaining an optimized image sequence;

[0032] Step S12, obtaining the original vibration signal collected by the vibration sensor; applying wavelet transform to the original vibration signal, decomposing the signal into different frequency components, identifying and removing high-frequency noise components, and then reconstructing the signal to obtain a vibration signal after noise reduction;

[0033] Step S13, obtaining the hopper position data recorded by the encoder; applying the Kalman filter algorithm to the hopper position data, combining the system dynamic model and the measurement data, estimating the real position state, filtering out random errors, and obtaining the smoothed hopper position data;

[0034] Step S14, obtaining the optimized image sequence, the vibration signal after noise reduction and the smoothed hopper position data; applying the timestamp alignment algorithm to them, aligning the data of different modes to a unified time axis according to the acquisition timestamp of each data, and obtaining a synchronized multi-modal data set;

[0035] Step S15, obtaining a synchronized multimodal data set; applying an improved SURF algorithm to the image sequence in the multimodal data set, extracting key points and descriptors in the image, and obtaining tablet features; applying a fast Fourier transform to the vibration signal, converting the time domain signal into a frequency domain representation, and obtaining frequency domain features; applying a differential algorithm to the hopper position data, calculating the position difference and second-order difference between adjacent time points, and obtaining velocity and acceleration features; combining the tablet features, frequency domain features, velocity and acceleration features, and obtaining a feature set;

[0036] Step S16, obtaining a feature set and a predefined basic parameter set; applying an adaptive parameter adjustment algorithm to the feature set, calculating and generating a set of initialized adjustable parameter values ​​according to the statistical characteristics of the features and the predefined basic parameters, and obtaining an initialized adjustable parameter set.

[0037] In this embodiment, through multimodal data acquisition and preprocessing, comprehensive acquisition and optimization processing of high-speed industrial camera, vibration sensor and encoder data are realized, providing high-quality input data for subsequent system modeling and control. Specifically, the adaptive exposure algorithm can dynamically adjust the camera parameters according to the scene brightness to ensure that clear images can be obtained under various lighting conditions; the wavelet denoising algorithm effectively removes high-frequency noise in the vibration signal and improves the signal-to-noise ratio of the signal; the Kalman filter algorithm smoothes the hopper position data and reduces the influence of random errors. These preprocessing steps significantly improve the data quality and lay a solid foundation for subsequent feature extraction and system modeling. The timestamp alignment algorithm ensures the synchronization of multimodal data, solves the problem of inconsistent sampling frequencies of different sensors, and enables the system to accurately capture the time correlation between various physical quantities. The improved SURF algorithm, fast Fourier transform and differential algorithm extract rich features from image, vibration and motion data respectively, which fully characterize the dynamic behavior of the system. Finally, the adaptive parameter adjustment algorithm dynamically initializes the system parameters according to the extracted features, improving the adaptability and robustness of the system. This series of technological innovations enables the system to quickly and accurately acquire and process multi-source heterogeneous data in complex and ever-changing working environments, laying a data foundation for achieving high-precision single-module follow-up control.

[0038] like Figure 3 As shown, the step S2 is specifically as follows:

[0039] Step S21, obtaining a feature set and an initialized adjustable parameter set; applying recursive least squares method to the feature set and the adjustable parameter set, updating the model parameter estimation, using a forgetting factor to give a higher weight to recent data, and identifying the dynamic characteristics of the system; by minimizing the weighted prediction error sum of squares, iteratively calculating the model parameters, and obtaining the nonlinear system model parameters;

[0040] Step S22, obtaining nonlinear system model parameters and a predefined set of candidate model structures; applying the Bayesian model selection algorithm to each candidate model structure, calculating the marginal likelihood of the model and combining it with the prior probability to obtain the posterior probability of each model; comparing the posterior probabilities of all candidate model structures, selecting the candidate model structure with the highest posterior probability, and obtaining an optimized nonlinear system model;

[0041] Step S23, obtaining a feature set and an optimized nonlinear system model; applying a delayed coordinate embedding method to calculate the mutual information function and determine the optimal time delay; and using a pseudo nearest neighbor algorithm to determine a suitable embedding dimension; performing a delayed coordinate transformation on the feature data according to the determined time delay and embedding dimension, reconstructing the system state space, and obtaining a reconstructed state vector;

[0042] Step S24, obtain the reconstructed state vector and the synchronized multimodal data set; apply the adaptive fault-tolerant Kalman filtering algorithm, first predict the state according to the system model; then use the multi-source measurement data to update the state; in this process, dynamically adjust the process noise covariance and the measurement noise covariance to adapt to the changes in the system state and measurement characteristics; identify and handle possible sensor failures or abnormal measurements by introducing a fault detection mechanism; obtain the estimated system state and the corresponding covariance matrix through iterative prediction and update steps.

[0043] In this embodiment, accurate description and real-time state tracking of complex dynamic systems are achieved through nonlinear system modeling and state estimation, providing reliable model support and state information for subsequent predictive control. Specifically, the combination of recursive least squares method and forgetting factor enables the system to adaptively identify time-varying parameters and effectively capture changes in the dynamic characteristics of the system; the Bayesian model selection algorithm selects the model structure that best reflects the essential characteristics of the system by evaluating the posterior probability of different model structures, thereby improving the generalization ability and prediction accuracy of the model. The delayed coordinate embedding method cleverly utilizes the historical information of the system, reconstructs the state space of the system by selecting appropriate embedding dimensions and time delays, and effectively solves the problem that the state cannot be directly measured. The adaptive fault-tolerant Kalman filter algorithm fuses the system model prediction with multi-source measurement data, which not only improves the accuracy of state estimation, but also enhances the system's robustness to sensor failures and abnormal measurements. Together, the system can accurately grasp the motion state of the hopper and the internal dynamics of the system in the scenario of single-module follow-up control, providing the necessary information support for achieving high-precision trajectory tracking and vibration suppression. Furthermore, the adaptive nature of the approach enables it to cope with slow drifts in system parameters and changes in the operating environment, ensuring long-term operational stability and reliability.

[0044] like Figure 4 As shown, the step S3 is specifically as follows:

[0045] Step S31, obtaining the estimated system state, the optimized nonlinear system model and the predefined control target; constructing the system state prediction equation according to the nonlinear system model; defining the cost function related to the control target, including the state deviation term, the control input term and the smoothing term; setting the constraints related to the system physical limitations and performance requirements; combining the cost function and the constraints, constructing the model predictive control optimization problem, and obtaining the model predictive optimization control framework;

[0046] Among them, the model prediction optimization control framework includes:

[0047] System state prediction equation: It is built based on the optimized nonlinear system model and is used to predict the state changes of the system in the future. In the scenario of single-module follow-up control, it includes predicting the future position, speed, acceleration and other states of the hopper.

[0048] Cost function: The cost function defines the objective to be optimized by the control system.

[0049] It includes the following parts:

[0050] State Deviation Term: Measures the difference between the actual state of the system and the desired state. In follow-up control, this may represent the deviation between the actual position of the bucket and the target trajectory.

[0051] Control input: Considers the size of the control signal, usually used to limit the amplitude of the control action to avoid overly drastic control.

[0052] Smoothing term: used to make the control signal smoother and reduce mutations. It is particularly important in hopper motion control and can reduce vibration and mechanical stress.

[0053] Constraints: Constraints reflect the physical limitations and performance requirements of the system. They include:

[0054] Upper and lower limits for the hopper position, maximum permissible speed and acceleration, actuator output limits, safety-related constraints such as minimum safe distances to other components.

[0055] Mathematical formulation of the optimization problem:

[0056] Combining all the above elements together forms a complete mathematical optimization problem. It is usually expressed as:

[0057] min J(x, u); st x(k+1) = f(x(k), u(k)); g(x, u)<= 0; h(x, u) = 0;

[0058] Among them, J is the cost function, x is the system state, u is the control input, f is the system dynamic equation, g and h are the inequality and equality constraints respectively.

[0059] Prediction horizon and control horizon: Although not directly defined in the problem statement, these two parameters have a significant impact on the size and complexity of the problem: the prediction horizon (Hp) defines the time span of the system state prediction, and the control horizon (Hc) defines the length of the optimized control sequence.

[0060] Step S32, obtaining the estimated system state, the optimized nonlinear system model and the current execution efficiency data; applying the dynamic time domain adjustment algorithm, calculating the system characteristic time constant according to the system model; and evaluating the computing resource load in combination with the current execution efficiency; dynamically adjusting the prediction time domain length according to the characteristic time constant and the computing resource load to ensure that the prediction covers the key system dynamics; adjusting the control time domain length to balance the control performance and the computational complexity; obtaining the optimized prediction time domain and control time domain through the iterative optimization process;

[0061] Step S33, obtaining the estimated system state, covariance matrix and predefined safety margin parameters; using the probabilistic constraint generation algorithm, constructing a state distribution model according to the mean and covariance of the state estimate; and determining the allowable probability of constraint violation based on the safety margin parameters; using the chance constraint theory to convert the deterministic constraints into probabilistic constraints; for nonlinear constraints, applying linearization or sampling methods for probabilistic processing; evaluating the feasibility of the probabilistic constraints by numerical integration or Monte Carlo method, and finally obtaining a robust control constraint set;

[0062] Step S34, obtain the model predictive optimization control framework, the optimized prediction time domain and control time domain, and the robust control constraint set; construct a multi-objective optimization problem, decompose the original cost function into multiple sub-objectives, including control performance, energy efficiency and constraint satisfaction; then introduce weight coefficients to form a weighted and multi-objective function; integrate the robust control constraint set into the optimization problem so that it satisfies the constraints under state uncertainty; adjust the dimension of the optimization problem according to the optimized prediction and control time domains; combine the multi-objective function, constraints and optimization variable range to obtain a complete optimization problem model.

[0063] In this embodiment, through the design of predictive control strategy, accurate prediction and optimization control of the future behavior of the system are achieved, and the accuracy and robustness of the single-module follow-up control are effectively improved. Specifically, the construction of the model predictive control problem integrates the dynamic characteristics, control objectives and constraints of the system into the optimization framework, so that the control strategy can achieve optimal performance while satisfying multiple constraints. The dynamic time domain adjustment algorithm dynamically optimizes the prediction and control time domain according to the system characteristics and execution efficiency, which not only ensures the accuracy of the prediction, but also balances the computational complexity, so that the control strategy can achieve real-time optimization under limited computing resources. The probabilistic constraint generation algorithm takes into account the uncertainty of state estimation, and enhances the robustness of the control strategy to model errors and external interference by converting deterministic constraints into probabilistic constraints. The construction of the multi-objective optimization problem comprehensively considers the control performance, robustness and constraint satisfaction, so that the system can find the best balance between multiple conflicting objectives. The control strategy can accurately predict the motion trajectory of the hopper in the scenario of single-module follow-up control, and optimize the control input according to the prediction results, effectively suppressing vibration and improving tracking accuracy. At the same time, the introduction of probabilistic constraints makes the control strategy more robust, able to cope with model errors, parameter uncertainties and external interference, and ensure that the system can maintain stable and reliable performance under various working conditions.

[0064] like Figure 5 As shown, the step S4 is specifically as follows:

[0065] Step S41, obtaining a complete optimization problem model and information on currently available computing resources of the system; applying an adaptive solver selection algorithm to analyze the characteristics of the optimization problem, including problem size, constraint type, and properties of the objective function; evaluating currently available computing resources, including processor performance, memory capacity, and parallel computing capabilities; selecting the most suitable optimization algorithm from a predefined solver library based on problem characteristics and resource conditions; configuring parameters of the selected solver, including convergence threshold, maximum number of iterations, and numerical precision, to obtain a configured solver;

[0066] Step S42, obtaining a complete optimization problem model and a configured solver; using a parallel computing framework, decomposing the optimization problem into multiple sub-problems; solving these sub-problems simultaneously on multiple processing cores; in the solving process, dynamically allocating computing resources to balance the computing load of each sub-problem; through iterative calculation and information exchange, gradually converging to a global optimal solution; obtaining an optimal control sequence, the optimal control sequence includes hopper speed v(t) and displacement d(t) control parameters for multiple future time steps;

[0067] Step S43, obtaining the optimal control sequence, the optimized nonlinear system model and the predefined uncertainty description; performing Monte Carlo simulation to generate a large number of random disturbance samples according to the uncertainty description; applying the optimal control sequence to the system model for each sample, simulating the system response and collecting all simulation results, and calculating statistical indicators, including average performance, worst-case performance and performance variance; calculating the robustness index of the control strategy based on the statistical indicators; obtaining sensitivity analysis results by analyzing the degree of influence of different uncertain factors on system performance;

[0068] Step S44, obtain the current control parameter u*(t) = {v(t), d(t)} in the optimal control sequence and the current control cycle parameter Tc of the system; execute the control action, and send the speed v(t) and displacement d(t) instructions to the hopper drive system; within the control cycle Tc, continuously monitor the actual motion state of the hopper and the system response; collect real-time sensor data, including the actual position, speed and acceleration information of the hopper; process the collected data, calculate the accuracy and response time of the control execution, and obtain the execution efficiency E(t); compare the actual executed hopper motion parameters with the control instructions, and obtain the actual executed hopper control parameters u_actual(t) = {v_actual(t), d_actual(t)}; integrate all monitoring and calculation results to obtain the complete actual system response Y(t).

[0069] In this embodiment, through adaptive optimization and control execution, efficient solution and robust execution of control strategy are achieved, which greatly improves the real-time and reliability of single-module follow-up control. Specifically, the adaptive solver selection algorithm dynamically selects the most suitable optimization algorithm according to the characteristics of the optimization problem and the available computing resources, ensuring that high-quality solutions are obtained within a limited computing time; the application of the parallel computing framework improves the speed of solving the optimization problem, enabling the system to complete complex optimization calculations in each control cycle and achieve true real-time optimization control. Monte Carlo simulation evaluates the robustness of the control strategy, and provides a probability distribution of the control effect by considering various uncertain factors, helping operators to better understand and manage control risks. The real-time monitoring and feedback mechanism enables the system to detect execution deviations in a timely manner and make adjustments, thereby improving the accuracy and reliability of control. Together, the system can quickly calculate the optimal control sequence in the scenario of single-module follow-up control, and dynamically adjust it during execution, effectively responding to various uncertainties and disturbances. Especially in scenarios with high-speed motion and precision control requirements, this method can significantly improve the tracking accuracy and motion smoothness of the hopper, reduce vibration and positioning errors, and improve production efficiency and product quality. At the same time, robustness assessment and real-time monitoring mechanisms enhance the reliability and security of the system, reduce operational risks, and provide guarantees for long-term stable operation.

[0070] According to one aspect of the present application, step S12 is specifically:

[0071] Step S121, obtaining an original vibration signal collected by a vibration sensor; applying an adaptive segmentation algorithm to the original vibration signal, dividing the signal into at least two time periods according to the local statistical characteristics of the signal, and obtaining a segmented vibration signal sequence;

[0072] Step S122, obtaining the segmented vibration signal sequence, applying an adaptive basis function selection algorithm to each signal segment, selecting the most suitable wavelet basis function from a predefined basis function library according to the frequency characteristics of the signal, and obtaining the optimal wavelet basis function corresponding to each signal segment;

[0073] Step S123, obtaining the segmented vibration signal sequence and the optimal wavelet basis function; applying the corresponding optimal wavelet basis function to each signal segment, performing multi-scale wavelet decomposition, and obtaining the wavelet coefficient of each signal segment;

[0074] Step S124, obtaining wavelet coefficients, and applying an adaptive threshold algorithm, calculating an optimal threshold according to the statistical distribution of the wavelet coefficients at each scale, performing soft threshold processing on the wavelet coefficients, and obtaining the denoised wavelet coefficients;

[0075] Step S125, obtaining the denoised wavelet coefficients and the optimal wavelet basis function, using inverse wavelet transform to reconstruct each signal segment, and then splicing all the reconstructed signal segments to obtain a complete denoised vibration signal;

[0076] Step S126, obtaining the vibration signal after noise reduction, applying an adaptive smoothing algorithm, dynamically adjusting the smoothing window size according to the local curvature characteristics of the signal, performing final smoothing processing on the signal, and obtaining an optimized noise reduction vibration signal.

[0077] In this embodiment, through advanced vibration data processing technology, the refined analysis and optimization processing of complex vibration signals are realized, the quality and information content of vibration data are significantly improved, and a more reliable vibration feature input is provided for single-module follow-up control. Specifically, the adaptive segmentation algorithm dynamically divides the time period according to the local statistical characteristics of the signal, effectively processes the non-stationary vibration signal, and captures the vibration characteristics of the system under different working conditions. The adaptive basis function selection algorithm selects the most suitable wavelet basis function for each signal segment, improves the efficiency and accuracy of the wavelet transform, and enables the system to more accurately separate the vibrations of different frequency components. Multi-scale wavelet decomposition realizes the time-frequency localization analysis of the signal and effectively identifies the vibration characteristics at different scales. The adaptive threshold algorithm performs soft threshold denoising in the wavelet domain, dynamically adjusts the threshold according to the statistical distribution of the wavelet coefficients at each scale, and effectively removes noise while retaining the useful signal to the greatest extent. The inverse wavelet transform and signal segment splicing reconstruct the complete denoised vibration signal, maintaining the continuity and physical meaning of the signal. Finally, the adaptive smoothing algorithm dynamically adjusts the smoothing window according to the local curvature characteristics of the signal, further optimizing the signal quality. This series of technical innovations enables the system to accurately capture and analyze various vibration modes during the hopper movement process, including structural vibration, external interference, and vibration caused by material flow, in the scenario of single-module follow-up control. By providing high-quality vibration characteristics, this method provides an important basis for subsequent system modeling and control optimization, helps to achieve more accurate vibration suppression and motion control, and improves the stability and reliability of the system.

[0078] According to one aspect of the present application, step S15 is specifically:

[0079] Step S151, obtaining an optimized image sequence in a synchronous multimodal data set, and applying an improved SURF algorithm to construct a scale space pyramid, and then using a Hessian matrix to detect feature points at each scale space layer to obtain a set of candidate feature points;

[0080] Step S152, obtaining a set of candidate feature points, applying an adaptive non-maximum suppression algorithm, dynamically adjusting the suppression window size according to the complexity of the local image structure, screening the most significant feature points, and obtaining an optimized feature point set;

[0081] Step S153, obtaining an optimized feature point set, constructing a rotation invariant descriptor for each feature point, calculating the main direction of the feature point neighborhood, and calculating the Haar wavelet response in the main direction coordinate system to obtain a feature point descriptor set;

[0082] Step S154, obtaining the denoised vibration signal in the synchronous multimodal data set, applying an adaptive time-frequency analysis algorithm, dynamically adjusting the time and frequency resolution according to the non-stationary characteristics of the signal, and obtaining a time-frequency representation of the signal;

[0083] Step S155, obtaining the obtained time-frequency representation, using a peak detection algorithm to identify significant peaks on the time-frequency plane, extracting the time, frequency and amplitude information of the significant peaks, and obtaining a feature vector of the vibration signal;

[0084] Step S156, obtaining the smoothed hopper position data in the synchronous multimodal data set, using an adaptive polynomial fitting algorithm, dynamically selecting the most suitable polynomial order according to the local complexity of the data, performing piecewise fitting on the position data, and obtaining a polynomial representation of the position;

[0085] Step S157, obtaining a position polynomial representation, calculating the first and second order derivatives of the polynomial, and obtaining velocity and acceleration estimates; applying a statistical feature extraction algorithm to the velocity and acceleration sequence, calculating statistics including mean, variance, and peak value, and obtaining a motion feature vector;

[0086] Step S158, obtain a feature point descriptor set, a vibration feature vector, and a motion feature vector, perform normalization processing, and use the weighted summation method in the feature fusion algorithm to combine the features of different modes, wherein the weight is dynamically adjusted according to the discriminative ability of each feature to obtain a fused multi-modal feature set.

[0087] In this embodiment, through multimodal feature extraction and fusion, a comprehensive characterization and efficient representation of the system state is achieved, providing a rich and compact feature set for single-mode follow-up control. Specifically, the improved SURF algorithm effectively captures the significant structural features in the image by constructing a scale space pyramid and using the Hessian matrix to detect feature points; the adaptive non-maximum suppression algorithm dynamically adjusts the suppression window according to the complexity of the local image structure, ensuring that the extracted feature points have strong representativeness and stability. The construction of the rotation invariant descriptor enhances the robustness of the feature, enabling the system to accurately identify and track the target when the hopper rotates and the viewing angle changes. The adaptive time-frequency analysis algorithm dynamically adjusts the time and frequency resolution for the non-stationary characteristics of the vibration signal, providing a high-quality time-frequency representation of the signal; the peak detection algorithm extracts significant features from the time-frequency plane, effectively compressing the representation of the vibration data. The adaptive polynomial fitting algorithm performs segmented fitting on the hopper position data, which not only smoothes the original data, but also provides a compact representation of the motion trend through the polynomial coefficients; the velocity and acceleration estimates obtained by calculating the polynomial derivatives, as well as the related statistical characteristics, comprehensively characterize the motion state of the hopper. Finally, the feature fusion algorithm combines the features of different modes organically through normalization and weighted summation to form a multimodal feature set that comprehensively reflects the system state. This enables the system to use visual, vibration and position information to comprehensively and accurately describe the motion state and surrounding environment of the hopper in the scenario of single-module follow-up control, providing high-quality input data for subsequent system modeling and control optimization. The fusion of multimodal features enhances the system's robustness to single-mode faults, improves the reliability and accuracy of state estimation, and lays a solid data foundation for achieving high-performance follow-up control.

[0088] According to one aspect of the present application, step S23 is specifically:

[0089] Step S231, obtaining a feature set and an optimized nonlinear system model, applying a mutual information algorithm to calculate the mutual information between the time series of each dimensional data in the feature set and its different time delayed versions, and obtaining a mutual information function curve;

[0090] Step S232, obtaining a mutual information function curve, applying an adaptive threshold selection algorithm to identify the first local minimum of the mutual information function, determining an optimal time delay τ, and obtaining a reconstructed time delay parameter;

[0091] Step S233, obtaining a feature set, an optimized nonlinear system model and an optimal time delay τ, applying an improved false nearest neighbor algorithm, sequentially increasing the embedding dimension, calculating the false nearest neighbor ratio under each dimension, until the false nearest neighbor ratio drops below a preset threshold, and obtaining the optimal embedding dimension m;

[0092] Step S234, obtaining a feature set, an optimal time delay τ and an optimal embedding dimension m, and constructing a delay coordinate matrix for each time series in the feature set, wherein each row of the matrix represents a reconstructed state vector, to obtain a preliminary reconstructed state space;

[0093] Step S235, obtaining a preliminary reconstructed state space and an optimized nonlinear system model; applying a principal component analysis algorithm to calculate the covariance matrix of the reconstructed state space, performing eigenvalue decomposition, selecting the first k principal components with the largest contribution rate, and obtaining a reduced-dimensional state space representation; k is a natural number;

[0094] Step S236, obtain the state space representation after dimensionality reduction, apply the local linear embedding algorithm to find the nearest neighbor point for each state point, construct local reconstruction weights, and then use these weights to reconstruct the data points in the low-dimensional space to obtain the final optimized reconstructed state vector.

[0095] In this embodiment, through the advanced state space reconstruction technology, effective modeling and state estimation of nonlinear dynamic systems are achieved, providing accurate system state representation for single-module follow-up control. Specifically, the mutual information algorithm determines the optimal time delay by analyzing the autocorrelation of the time series, effectively capturing the intrinsic dynamic characteristics of the system; the adaptive threshold selection algorithm intelligently identifies the local minimum on the mutual information function curve to ensure the optimal selection of the time delay parameter. The improved pseudo nearest neighbor algorithm determines the optimal embedding dimension that can fully unfold the system dynamics by gradually increasing the embedding dimension and evaluating the pseudo nearest neighbor ratio, avoiding the increase in computational complexity and overfitting problems caused by excessive dimensionality. The construction of the delay coordinate matrix converts the original time series into a high-dimensional state vector, effectively reconstructs the phase space of the system, and reveals the potential dynamic structure of the system. The principal component analysis (PCA) algorithm extracts the most representative principal components by analyzing the covariance structure of the reconstructed state space, realizes the dimensionality reduction of the state space, retains key dynamic information, and reduces the computational complexity of subsequent processing. The local linear embedding (LLE) algorithm further optimizes the representation of the state space by maintaining the local geometric relationship between data points, and effectively handles the complex manifold structure of the nonlinear system. This series of technical innovations enables the system to accurately capture the nonlinear dynamic characteristics of the hopper motion, including possible chaotic behavior, periodic motion, and transient response, in the scenario of single-module follow-up control. The reconstructed state space provides an ideal working foundation for subsequent nonlinear prediction and control, enabling the system to better predict the motion trajectory of the hopper and achieve more accurate trajectory tracking and vibration suppression. At the same time, the reduced-dimensional state representation greatly reduces the computational complexity, enabling the system to achieve real-time control with limited computing resources, and improving the response speed and efficiency of the control system.

[0096] According to one aspect of the present application, step S24 is specifically:

[0097] Step S241, obtaining the reconstructed state vector and the synchronized multimodal data set, applying an adaptive noise estimation algorithm, utilizing the statistical characteristics of the innovation sequence, dynamically estimating the process noise covariance Q(k) and the measurement noise covariance R(k), and obtaining an adaptive noise model;

[0098] Step S242, obtaining an adaptive noise model, reconstructing a state vector and a synchronous multimodal data set, applying a strong tracking filtering algorithm, introducing an attenuation factor, dynamically adjusting the prediction error covariance matrix, enhancing the filter's adaptability to model mismatch, and obtaining a corrected prediction error covariance;

[0099] Step S243, obtaining the corrected prediction error covariance, the reconstructed state vector and the synchronized multimodal data set, applying the adaptive factor method, calculating the adaptive factor according to the statistical characteristics of the new information sequence, weighting the contribution of the measurement update, and obtaining the adaptive gain matrix;

[0100] Step S244, obtaining an adaptive gain matrix, a reconstructed state vector and a synchronized multimodal data set, executing a prediction step of a Kalman filter, using a system model to make a one-step prediction of the state, and obtaining a priori state estimation and error covariance;

[0101] Step S245, obtaining a priori state estimation and error covariance, an adaptive gain matrix and a synchronous multimodal data set; executing a Kalman filter update step, correcting the a priori estimation in combination with the measurement data, and obtaining a posteriori state estimation and error covariance;

[0102] Step S246, obtaining a posteriori state estimation and error covariance, applying a fault detection and isolation algorithm to calculate a normalized innovation sequence, and performing hypothesis testing to identify possible sensor faults or abnormal measurements to obtain a fault diagnosis result;

[0103] Step S247, obtain the a posteriori state estimate, error covariance, and fault diagnosis results; apply the robust data fusion algorithm, adjust the credibility of each sensor data according to the fault diagnosis results, recalculate the fused state estimate and uncertainty, and obtain the final estimated system state and covariance matrix.

[0104] In this embodiment, the adaptive fault-tolerant Kalman filtering technology is used to achieve high-precision estimation and robust tracking of the state of the nonlinear system, significantly improving the accuracy and reliability of the single-module follow-up control. Specifically, the adaptive noise estimation algorithm analyzes the statistical characteristics of the innovation sequence and dynamically adjusts the covariance matrix of the process noise and the measurement noise, so that the filter can adaptively respond to changes in the system noise characteristics, thereby improving the accuracy and stability of the state estimation. The strong tracking filter algorithm introduces an attenuation factor to dynamically adjust the prediction error covariance matrix, enhances the filter's adaptability to model mismatch, and enables the system to maintain good tracking performance when the parameters change slowly. The adaptive factor method dynamically adjusts the filter gain according to the statistical characteristics of the new information sequence, optimizes the weight distribution of the measurement update when the reliability of the measurement data is uncertain, and improves the robustness of the filter. The execution of the prediction step and the update step realizes the one-step prediction and measurement correction of the system state, providing an accurate state estimation for the control system. The fault detection and isolation (FDI) algorithm analyzes the normalized innovation sequence and detects possible sensor failures or abnormal measurements in real time, thereby enhancing the fault tolerance of the system. The robust data fusion algorithm dynamically adjusts the credibility of each sensor data according to the fault diagnosis results, realizes the intelligent fusion of multi-source information, and further improves the reliability and accuracy of state estimation. It enables the system to accurately estimate key states such as the position, velocity, and acceleration of the hopper in complex scenarios of single-module follow-up control, and maintain high-precision state tracking even in the presence of sensor noise, model errors, and potential faults. By providing reliable state estimation, this method provides a solid foundation for subsequent predictive control, enabling the system to more accurately predict and control the motion trajectory of the hopper, effectively suppress vibration, and improve positioning accuracy and motion smoothness. At the same time, its adaptive and fault-tolerant characteristics significantly enhance the robustness and reliability of the system, enabling it to maintain stable operation under various complex working conditions and reduce the risk of control failure due to state estimation errors.

[0105] According to one aspect of the present application, step S32 is specifically:

[0106] Step S321, obtain the estimated system state, the optimized nonlinear system model and the current execution efficiency data. Apply the system characteristic time scale analysis algorithm to calculate the main dynamic modes of the system and their corresponding characteristic time constants to obtain the multi-scale dynamic characteristics description of the system.

[0107] Step S322: Obtain multi-scale dynamic characteristics description and current execution efficiency data. Apply a computational complexity evaluation algorithm to estimate the computation time under different prediction time domain lengths according to the complexity of the system model and available computing resources, and obtain a relationship curve between the computation time and the prediction time domain.

[0108] Step S323, obtain the relationship curve between calculation time and prediction time domain and multi-scale dynamic characteristics description. Apply the multi-objective optimization algorithm, take the prediction accuracy and calculation efficiency as the optimization goals, find the best prediction time domain length, and obtain the initially optimized prediction time domain Hp.

[0109] Step S324, obtain the preliminary optimized prediction time domain Hp, estimate the system state and the optimized nonlinear system model. Apply the prediction error analysis algorithm, perform multi-step predictions at different prediction lengths, calculate the changing trend of the prediction error as the time domain increases, and obtain the prediction reliability evaluation result.

[0110] Step S325, obtain the prediction reliability evaluation result and the preliminary optimization prediction time domain Hp. Apply the adaptive time domain adjustment algorithm, fine-tune the prediction time domain length according to the change trend of the prediction reliability, balance the prediction accuracy and calculation efficiency, and obtain the final optimized prediction time domain Hp*.

[0111] Step S326, obtain the final optimized prediction time domain Hp* and the optimized nonlinear system model. Apply the control complexity analysis algorithm to evaluate the impact of different control time domain lengths on system performance and computing load, and obtain the control complexity evaluation result.

[0112] Step S327, obtain the control complexity evaluation result and the final optimized prediction time domain Hp*. Apply the adaptive control time domain optimization algorithm, select the shortest effective control time domain under the premise of ensuring the control performance, and obtain the optimized control time domain Hc*.

[0113] In this embodiment, the dynamic optimization of the predictive control strategy is realized by the adaptive prediction time domain adjustment technology, which significantly improves the real-time performance and control performance of the single-module follow-up control. Specifically, the system characteristic time scale analysis algorithm provides a theoretical basis for the selection of the prediction time domain by identifying the main dynamic modes of the system and their corresponding characteristic time constants, ensuring that the prediction can cover the key dynamic behaviors of the system. The computational complexity evaluation algorithm establishes a relationship model between the prediction time domain length and the calculation time by analyzing the system model complexity and the available computing resources, and provides constraints on the computational efficiency for the time domain optimization. The multi-objective optimization algorithm finds the best balance between prediction accuracy and computational efficiency, preliminarily determines the optimized prediction time domain, and provides a good starting point for subsequent fine-tuning. The prediction error analysis algorithm quantifies the changing trend of the prediction reliability over time by simulating the system response under different prediction lengths, and provides accurate performance indicators for further optimization of the prediction time domain. The adaptive time domain adjustment algorithm fine-tunes the prediction time domain according to the changing trend of the prediction reliability, maximizing the computational efficiency while ensuring the prediction accuracy. The control complexity analysis algorithm evaluates the impact of the control time domain length on system performance and computational load, and provides a basis for the optimization of the control time domain. The adaptive control time domain optimization algorithm selects the shortest effective control time domain under the premise of ensuring control performance, further improving the computational efficiency of the control strategy. This enables the system to adaptively adjust the prediction and control time domains according to the complexity of the hopper movement and changes in the system's computing power in the dynamic scenario of single-module follow-up control. When the hopper movement is relatively simple, the system can select a shorter time domain to improve the control frequency and response speed; when the hopper movement is complex or there is a long-term trend, the system can select a longer time domain to improve prediction accuracy and control stability. This dynamic adjustment mechanism not only optimizes the control performance, but also significantly improves the utilization efficiency of computing resources, enabling the system to achieve high-performance real-time control under limited computing power. At the same time, by timely adjusting the prediction and control strategies, this method enhances the system's adaptability to external interference and changes in operating conditions, and improves the robustness and reliability of follow-up control.

[0114] According to one aspect of the present application, step S33 is specifically:

[0115] Step S331, obtain the estimated system state, covariance matrix and predefined safety margin parameters. Apply the probability distribution fitting algorithm, fit the probability distribution model of the state variable according to the mean and covariance of the state estimate, and obtain the probability representation of the state uncertainty.

[0116] Step S332: Obtain the state uncertainty probability representation and predefined safety margin parameters. Apply the chance constraint conversion algorithm to convert the deterministic constraint into a probability form, set the allowed probability threshold of constraint violation, and obtain a preliminary probability constraint expression.

[0117] Step S333, obtaining a preliminary probability constraint expression and an optimized nonlinear system model. For nonlinear constraints, a linearized approximation algorithm is applied to perform Taylor expansion near the current working point to obtain a linearized probability constraint expression.

[0118] Step S334: Obtain a linearized probability constraint expression. Apply an analytical probability integral algorithm to analytically solve the simple form of probability constraint to obtain an equivalent deterministic constraint expression.

[0119] Step S335: Obtain a complex probability constraint expression that cannot be solved analytically. Apply the Monte Carlo sampling algorithm to generate a large number of random samples that obey the state distribution, evaluate the probability of constraint satisfaction, and obtain a probability estimation result based on sampling.

[0120] Step S336: Obtain equivalent deterministic constraints and obtain sampling-based probability estimation results. Apply an adaptive constraint shrinkage algorithm to conservatively adjust the constraint boundaries according to the probability estimation results to ensure that the probability requirements of the original constraints are met while considering the uncertainty, and obtain a shrunken robust constraint set.

[0121] Step S337, obtaining the post-shrinkage robust constraint set. Apply the constraint feasibility verification algorithm, verify the feasibility of the post-shrinkage constraint set through local optimization or feasibility sampling method, and make fine adjustments if necessary to obtain the final robust control constraint set.

[0122] In this embodiment, the control constraint design considering uncertainty is realized through the robust constraint generation technology, which significantly improves the robustness and safety of the single-module follow-up control. Specifically, the probability distribution fitting algorithm constructs a probability distribution model of the state variable according to the statistical characteristics of the state estimation, which provides a theoretical basis for the subsequent probability constraint processing. The opportunity constraint conversion algorithm converts the deterministic constraint into a probabilistic form, and by setting the allowable probability threshold of violating the constraint, it realizes the moderate relaxation of the constraint conditions under the premise of ensuring the safety of the system, and increases the flexibility of the control strategy. For nonlinear constraints, the linearization approximation algorithm obtains a linearized probability constraint expression by Taylor expansion near the current working point, which simplifies the subsequent probability integral calculation. The analytical probability integral algorithm accurately solves the simple form of probability constraints and obtains an equivalent deterministic constraint expression, which improves the computational efficiency of constraint processing. For complex probability constraints, the Monte Carlo sampling algorithm generates a large number of random samples to evaluate the probability of constraint satisfaction, provides a numerical approximate solution, and ensures the universality of constraint processing. The adaptive constraint shrinkage algorithm conservatively adjusts the constraint boundary according to the probability estimation result to ensure that the probability requirements of the original constraint are met under the consideration of uncertainty, and enhances the conservatism and safety of the control strategy. The constraint feasibility verification algorithm verifies and fine-tunes the contracted constraint set through local optimization or feasibility sampling methods, ensuring the feasibility and effectiveness of the final generated robust control constraint set. This enables the system to generate robust control constraints that take into account state estimation errors, model uncertainties, and external disturbances in the uncertain environment of single-module follow-up control. During the hopper movement process, even in the presence of unpredictable disturbances or parameter drifts, the system can remain within the safe operating range, effectively avoiding constraint violations and potential safety risks. At the same time, through the introduction of probabilistic constraints, the control strategy retains a certain degree of flexibility while ensuring safety, avoiding overly conservative control behavior, and achieving a good balance between safety and control performance. This robust constraint generation method significantly improves the reliability and adaptability of the follow-up control system, enabling it to maintain stable and efficient operation in various complex and changing working environments.

[0123] According to one aspect of the present application, step S42 is specifically:

[0124] Step S421, obtain the complete optimization problem definition and the configured solver. Apply the problem decomposition algorithm to decompose the large-scale optimization problem into multiple sub-problems, establish the association relationship between the sub-problems, and obtain the problem decomposition structure.

[0125] Step S422: Obtain the problem decomposition structure. Apply the load balancing algorithm, formulate a task allocation strategy according to the computational complexity of each sub-problem and the available computing resources, and obtain an optimized task allocation solution.

[0126] Step S423, obtaining the task allocation scheme and the post-configuration solver. Applying the parallel solver initialization algorithm, initializing the solver instances on multiple processing cores, setting the initial parameters and constraints of each instance, and obtaining an initialized parallel solver set.

[0127] Step S424: Obtain a set of parallel solvers. Apply a distributed optimization algorithm to solve sub-problems on multiple processing cores at the same time, exchange boundary information and intermediate results regularly, and obtain a local optimal solution for each sub-problem.

[0128] Step S425: Obtain a local optimal solution set. Apply a de-aggregation algorithm to synthesize the local optimal solutions of each sub-problem, coordinate the coupling relationship between the sub-problems, and obtain a globally consistent solution.

[0129] Step S426, obtaining a global consistent solution. Apply a convergence analysis algorithm to evaluate the convergence degree of the global solution. If the convergence standard is not met, return to S424 to continue iterating, otherwise output the current solution as the final solution to obtain the optimal control sequence.

[0130] Step S427, obtaining the optimal control sequence. Applying the control parameter extraction algorithm, extracting the hopper speed v(t) and displacement d(t) control parameters at the current moment and in the next few time steps from the optimal control sequence, and obtaining the final hopper control parameter sequence.

[0131] In this embodiment, through the parallel optimization computing technology, the efficient solution of large-scale predictive control problems is achieved, and the real-time performance and optimization quality of single-module follow-up control are significantly improved. Specifically, the problem decomposition algorithm decomposes the complex optimization problem into multiple interrelated sub-problems, reduces the scale and complexity of a single problem, and lays the foundation for parallel computing. The load balancing algorithm formulates the optimal task allocation plan according to the computational complexity of the sub-problems and the available computing resources, ensuring the efficient use of computing resources. The parallel solver initialization algorithm sets up solver instances on multiple processing cores at the same time, improving the parallelism of the calculation. The distributed optimization algorithm achieves rapid convergence of the global optimal solution by executing sub-problem solving on multiple processing cores at the same time and regularly exchanging boundary information and intermediate results. The deaggregation algorithm effectively coordinates the coupling relationship between sub-problems, ensuring that the local optimal solution can be combined into a globally consistent solution. The convergence analysis algorithm dynamically adjusts the iterative process by evaluating the convergence degree of the global solution, minimizing the calculation time while ensuring the quality of the solution. The control parameter extraction algorithm extracts key hopper speed and displacement control parameters from the optimal control sequence, providing directly available instructions for actual control execution. This enables the system to quickly solve complex predictive control optimization problems in high-speed dynamic scenarios of single-module follow-up control. Through parallel computing, the system can complete large-scale optimization calculations in each control cycle to achieve true real-time optimization control. This not only improves the update frequency of the control strategy and enhances the system's adaptability to rapidly changing working conditions, but also improves the global optimality of the control strategy by solving more complex and longer time-domain optimization problems. In scenarios such as high-speed hopper movement or complex trajectory tracking, this method can quickly generate optimal control sequences that take long-term goals into account, significantly improving trajectory tracking accuracy and motion smoothness. At the same time, the introduction of parallel computing greatly improves the utilization efficiency of computing resources, enabling the system to achieve high-performance control under limited hardware conditions and reducing implementation costs. By providing fast and high-quality optimization results, this method provides key technical support for achieving high-precision and high-reliability single-module follow-up control.

[0132] According to one aspect of the present application, step S43 is specifically:

[0133] Step S431, obtaining the optimal control sequence, the optimized nonlinear system model and the predefined uncertainty description. Applying the uncertainty modeling algorithm, constructing the probability model of parameter uncertainty and external interference according to the uncertainty description, and obtaining the random representation of the system uncertainty.

[0134] Step S432: Obtain random representation of system uncertainty. Apply the Latin hypercube sampling algorithm to generate a representative sample set in the uncertainty space, ensure that the samples cover all areas of the uncertainty space, and obtain an uncertainty sampling set.

[0135] Step S433: Obtain uncertainty sampling set, optimal control sequence and optimized nonlinear system model. Apply parallel Monte Carlo simulation algorithm to perform system simulation on multiple processing cores for each uncertainty sample to obtain a large number of system response trajectories.

[0136] Step S434: Obtain a set of system response trajectories. Apply statistical analysis algorithms to calculate statistical characteristics of system performance indicators, including mean, variance, worst-case performance, etc., to obtain performance statistics results.

[0137] Step S435: Obtain performance statistics results. Apply a robustness index calculation algorithm to calculate the probability that the system meets the performance requirements based on the performance statistics results and predefined performance requirements, and obtain a robustness index.

[0138] Step S436: Obtain a set of system response trajectories and a random representation of system uncertainty. Apply a sensitivity analysis algorithm to evaluate the impact of different uncertain factors on system performance, identify key uncertain parameters, and obtain sensitivity analysis results.

[0139] Step S437: Obtain robustness indicators and sensitivity analysis results. Apply a comprehensive evaluation algorithm, combine robustness indicators and sensitivity analysis results, generate a detailed robustness evaluation report, including the expected performance and potential risks of the system under uncertainty, and obtain the final robustness analysis results.

[0140] In this embodiment, through the robustness analysis technology, a comprehensive evaluation and optimization of the control strategy performance is achieved, and the reliability and adaptability of the single-module follow-up control are significantly improved. Specifically, the uncertainty modeling algorithm provides a mathematical description of the randomness of the system behavior by constructing a probabilistic model of parameter uncertainty and external interference, and lays a theoretical foundation for the subsequent robustness analysis. The Latin hypercube sampling algorithm generates a representative sample set in the uncertainty space, ensures the uniformity and comprehensiveness of the sampling, and improves the reliability of the subsequent simulation results. The parallel Monte Carlo simulation algorithm improves the simulation efficiency by executing a large number of system simulations on multiple processing cores at the same time, so that the system can evaluate the performance of a large number of uncertainties in a limited time. The statistical analysis algorithm provides the expected performance and risk assessment of the control strategy in an uncertain environment by calculating the statistical characteristics of the system performance indicators. The robustness index calculation algorithm combines the performance statistics with the predefined performance requirements, quantifies the probability that the system meets the performance requirements, and provides an intuitive metric for the evaluation of the control strategy. The sensitivity analysis algorithm identifies the key uncertain parameters by evaluating the degree of influence of different uncertain factors on the system performance, and provides guidance for the targeted optimization of the control strategy. The comprehensive evaluation algorithm combines the robustness index and the sensitivity analysis results to generate a detailed robustness evaluation report, which fully reflects the expected performance and potential risks of the system under uncertainty. It enables the system to comprehensively evaluate the robustness and reliability of the control strategy in complex scenarios of single-module follow-up control. By simulating a large number of possible uncertainty situations, the system can foresee potential performance risks and optimize the control strategy accordingly, thereby improving the fault tolerance and adaptability of the control system. During the movement of the hopper, the system can maintain stable and reliable performance even in the face of parameter drift, external interference or changes in working conditions. The application of sensitivity analysis results enables the control strategy to be optimized for the most critical uncertainties, improving the efficiency and effectiveness of optimization. By providing a detailed robustness evaluation report, this method not only provides a basis for the optimization of the control strategy, but also provides the system operator with an important tool for risk management, which helps to improve the reliability and safety of the entire system.

[0141] According to another aspect of the present application, a single-module follow-up control system is also provided, comprising:

[0142] at least one processor; and,

[0143] a memory communicatively connected to at least one of the processors; wherein,

[0144] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the single-module follow-up control method described in any of the above technical solutions.

[0145] In another embodiment of the present application,

[0146] Step S1: Obtain the original image sequence captured by the high-speed industrial camera, the original vibration signal collected by the vibration sensor, and the hopper position data recorded by the encoder. Apply the adaptive exposure algorithm to the original image sequence, adjust the image brightness and contrast, and obtain the optimized image sequence. Use the wavelet denoising algorithm to remove high-frequency noise on the original vibration signal to obtain the denoised vibration signal. Apply the Kalman filter algorithm to the hopper position data to smooth the position data and obtain the smoothed hopper position data. Input the optimized image sequence, the denoised vibration signal, and the smoothed hopper position data into the timestamp alignment algorithm to synchronize the multimodal data to obtain a synchronized multimodal data set. Apply the improved SURF algorithm to the image sequence in the synchronized multimodal data set to extract tablet features, apply the fast Fourier transform to the vibration signal to extract frequency domain features, and apply the differential algorithm to the hopper position data to calculate the velocity and acceleration features to obtain a feature set. According to the feature set and the predefined basic parameter set, the adaptive parameter adjustment algorithm is applied to initialize the adjustable parameters to obtain the initialized adjustable parameter set.

[0147] Step S2, obtain the feature set obtained in step S1 and the initialized adjustable parameter set. Apply recursive least squares method and forgetting factor to the feature set and the adjustable parameter set to identify the dynamic characteristics of the system and obtain the nonlinear system model parameters. According to the nonlinear system model parameters and the predefined model structure set, use the Bayesian model selection algorithm to evaluate the posterior probability of different model structures, select the optimal model structure, and obtain the optimized nonlinear system model. Obtain the feature set and the optimized nonlinear system model, apply the delayed coordinate embedding method, select the appropriate embedding dimension and time delay, reconstruct the system state space, and obtain the reconstructed state vector. Obtain the reconstructed state vector and the synchronized multimodal data set in step S1, apply the adaptive fault-tolerant Kalman filtering algorithm, combine the system model prediction and multi-source measurement data, estimate the system state, and obtain the estimated system state and covariance matrix.

[0148] Step S3, obtain the estimated system state obtained in step S2, the optimized nonlinear system model and the predefined control objective. Construct a model predictive control optimization problem, including defining the cost function and constraints, and obtain the model predictive control problem statement. Obtain the estimated system state, the optimized nonlinear system model and the current execution efficiency, apply the dynamic time domain adjustment algorithm, optimize the prediction and control time domains according to the system dynamic characteristics and execution efficiency, and obtain the optimized prediction time domain and control time domain. Obtain the estimated system state, covariance matrix and predefined safety margin, use the probabilistic constraint generation algorithm, consider the uncertainty of state estimation, generate probabilistic constraints, and obtain a robust control constraint set. Obtain the model predictive control problem statement, the optimized prediction time domain, control time domain and robust control constraint set, construct a multi-objective optimization problem, comprehensively consider the control performance, robustness and constraint satisfaction, and obtain a complete optimization problem definition.

[0149] Step S4, obtain the complete optimization problem definition obtained in step S3 and the computing resource information currently available in the system. Apply the adaptive solver selection algorithm, select the most suitable optimization algorithm according to the optimization problem characteristics and computing resource conditions, and obtain the configured solver. Obtain the complete optimization problem definition and the configured solver, use the parallel computing framework to solve the optimization problem, and obtain the optimal control sequence, including the speed and displacement control parameters of the hopper. Obtain the optimal control sequence, the optimized nonlinear system model obtained in step S2, and the predefined uncertainty description, perform Monte Carlo simulation, evaluate the robustness of the control strategy, and obtain robustness indicators and sensitivity analysis results. Obtain the optimal control sequence and the current control cycle parameters, execute control actions, control the hopper movement, monitor the system response, and obtain the actual system response, execution efficiency, and the actual executed hopper control parameters.

[0150] In the above embodiment, the relevant models and algorithms are specifically as follows:

[0151] The models and algorithms in multimodal data acquisition and preprocessing are as follows:

[0152] Let I(t) be the image data at time t, V(t) be the vibration signal, and P(t) be the hopper position data.

[0153] Adaptive exposure algorithm: I'(t) = f_AE(I(t), α, β); where f_AE is the adaptive exposure function, α is the brightness adjustment parameter, and β is the contrast adjustment parameter.

[0154] Wavelet denoising: V'(t) = W (-1) (T_λ(W(V(t))))); where W is the wavelet transform, W (-1) is the inverse wavelet transform, T_λ is the threshold function, and λ is the adaptive threshold.

[0155] Kalman filter smoothing: P'(t) = KF(P(t), Q, R); where KF is the Kalman filter function, Q is the process noise covariance, and R is the measurement noise covariance.

[0156] Timestamp alignment: D(t) = Align(I'(t), V'(t), P'(t)); where Align is the time alignment function and D(t) is the synchronized multimodal dataset.

[0157] Nonlinear system modeling and state estimation, the models and algorithms are as follows:

[0158] Recursive least squares method: θ(t) = θ(t-1) + K(t)(y(t) - φ T (t)θ(t-1));K(t) = P(t-1)φ(t)(λ + φ T (t)P(t-1)φ(t)) (-1) ;P(t) = (I - K(t)φ T (t))P(t-1) / λ;

[0159] Among them, θ is the model parameter, K is the gain matrix, y is the system output, φ is the regression vector, P is the covariance matrix, and λ is the forgetting factor.

[0160] Bayesian model selection: p(M_i|D) ∝ p(D|M_i)p(M_i); where Mi is the candidate model, D is the data, and p(M_i|D) is the posterior probability of the model.

[0161] Delayed coordinate embedding: X(t) = [x(t), x(t-τ), ..., x(t-(m-1)τ)]; where τ is the time delay and m is the embedding dimension.

[0162] Adaptive fault-tolerant Kalman filter: x*(t|t-1) = f(x*(t-1|t-1), u(t-1)); P(t|t-1) = F(t)P(t-1|t-1)F T (t) + Q(t); K(t) = P(t|t-1)H T (t)(H(t)P(t|t-1)H T (t) + R(t)) (-1) ;x*(t|t) = x*(t|t-1) + K(t)(z(t) - h(x*(t|t-1)));P(t|t) = (I - K(t)H(t))P(t|t-1);

[0163] Where x* is the state estimate, P is the estimation error covariance, K is the Kalman gain, z is the measurement, f and h are nonlinear functions, and F and H are the corresponding Jacobian matrices.

[0164] The predictive control strategy design, model and algorithm are as follows:

[0165] Model predictive control problem statement:

[0166] min J = Σ(||x(k+i|k) - x_ref(k+i|k)||_Q 2 + ||u(k+i|k)||_R 2 );st x(k+i+1|k) = f(x(k+i|k), u(k+i|k));g(x(k+i|k), u(k+i|k)) ≤ 0;i = 0, 1, ..., N-1;where J is the cost function, x is the system state, u is the control input, x_ref is the reference trajectory, Q and R are weight matrices, f is the system model, g is the constraint function, and N is the prediction time domain.

[0167] Dynamic time domain adjustment: N_opt = arg min E[J(N)]; st T_comp(N) ≤ T_max; where N_opt is the optimal prediction time domain, E[J(N)] is the expected cost, T_comp is the calculation time, and T_max is the maximum allowed calculation time.

[0168] Probabilistic constraint generation: Pr{g(x, u) ≤ 0} ≥ 1 – ε; where Pr represents the probability and ε is the allowed probability of violating the constraint.

[0169] The algorithms and models used in adaptive optimization and control execution are as follows:

[0170] Parallel optimization solution: u* = arg min J(x, u); st constraints.

[0171] Use a distributed optimization algorithm such as the alternating direction method of multipliers (ADMM) to solve.

[0172] Monte Carlo simulation: R = 1 / M Σ r(x_i, u*); where R is the robustness index, M is the number of simulations, r is the result of a single simulation, and x_i is the randomly generated initial state.

[0173] Control execution: u_actual(t) = u*(t) + K_fb(x(t) – x*(t)); where u_actual is the actual executed control input and K_fb is the feedback gain matrix.

[0174] In another embodiment of the present application, the relevant algorithm is as follows:

[0175] Hilbert-Huang Transform (HHT): HHT(V(t)) = H(IMF_i(t)); where: V(t): original vibration signal, IMF_i(t): the ith intrinsic mode function (Intrinsic Mode Function); H(): Hilbert transform; HHT(): Hilbert-Huang transform. It is used to analyze nonlinear and non-stationary signals, and is particularly suitable for processing complex vibration data.

[0176] Empirical Mode Decomposition (EMD): V(t) = Σ IMF_i(t) + r_n(t); where: V(t): original vibration signal, IMF_i(t) is the i-th intrinsic mode function; r_n(t) is the remaining signal (residual); Σ is the summation symbol. It is used to decompose complex signals into a finite number of intrinsic mode functions, which helps to analyze the intrinsic characteristics of the signal.

[0177] Wavelet Packet Transform (WPT): WPT(V(t)) = {d_j,k(t) | j = 1,...,J; k = 0,...,2^j-1}

[0178] Among them, V(t) is the original signal, d_j,k(t) is the j-th layer and k-th wavelet packet coefficient; J is the maximum number of decomposition layers, which provides a finer frequency resolution than ordinary wavelet transform and is suitable for extracting the characteristics of complex signals.

[0179] Singular Spectrum Analysis (SSA): X = [X_1 : ... : X_K]; [U, Σ, V] = SVD(X);

[0180] Where: X: trajectory matrix, Xi: delay vector, SVD(): singular value decomposition, U, Σ, V: matrix obtained by SVD decomposition, used to extract trends, periodic components and noise in the signal, which is very helpful for the analysis of complex time series.

[0181] Delay vector method: x(t) = [s(t), s(t+τ), ..., s(t+(m-1)τ)];

[0182] Where: x(t) is the reconstructed state vector, s(t) is the original time series, τ is the time delay, and m is the embedding dimension; by selecting appropriate time delay and embedding dimension, a multidimensional state space can be reconstructed from a univariate time series.

[0183] The mutual information method is used to select the optimal time delay: I(τ) = Σ p(s(t), s(t+τ)) log(p(s(t), s(t+τ)) / (p(s(t))p(s(t+τ))));

[0184] Where: I(τ) is the mutual information function, p() is the probability density function, s(t) is the original time series, and τ is the time delay; the first local minimum of I(τ) is selected as the optimal time delay.

[0185] The false nearest neighbor method (FNN) is used to determine the embedding dimension: R_d = |x_d+1(i) - x_d+1(j)| / |x_d(i)- x_d(j)|;

[0186] Where: R_d is the distance ratio, x_d(i) is the point i in the d-dimensional space; x_d(j) is the nearest neighbor of point i in the d-dimensional space. When R_d is greater than a certain threshold, it is considered a false nearest neighbor. The minimum dimension that reduces the false nearest neighbor ratio to a sufficiently low level is selected as the embedding dimension.

[0187] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. Single module follow-up control method, It is characterized in that The steps include: Step S1, obtaining an original image sequence captured by a high-speed industrial camera, an original vibration signal collected by a vibration sensor, and hopper position data recorded by an encoder, and preprocessing them respectively to obtain a synchronized multimodal data set; extracting a feature set from the multimodal data set, and applying an adaptive parameter adjustment algorithm to initialize adjustable parameters according to the feature set and a predefined basic parameter set to obtain an initialized adjustable parameter set; Step S2, based on the feature set and the adjustable parameter set, identify the system dynamic characteristics, obtain nonlinear system model parameters, build and optimize the nonlinear system model according to the nonlinear system model parameters, reconstruct the system state space based on the nonlinear system model, obtain the reconstructed state vector, combine the system model prediction and multi-source measurement data, estimate the system state, and obtain the estimated system state and covariance matrix; Step S3, based on the estimated system state, the optimized nonlinear system model and the predefined control objectives, a model predictive control optimization problem is constructed; a dynamic time domain adjustment algorithm is applied to optimize the prediction and control time domains, and a probabilistic constraint generation algorithm is used to generate a robust control constraint set; a multi-objective optimization problem is constructed by comprehensively considering control performance, robustness and constraint satisfaction, and a complete optimization problem model is obtained; Step S4: According to the complete optimization problem model and the computing resource information currently available in the system, an adaptive solver selection algorithm is used and a parallel computing framework is used to solve the optimization problem, an optimal control sequence is obtained, and the optimal control sequence is sent to each module; Conduct Monte Carlo simulation to evaluate the robustness of the control strategy; execute control actions, control the hopper movement, monitor the system response, and obtain the actual system response, execution efficiency, and actual executed hopper control parameters; The step S3 is specifically as follows: Step S31, obtaining the estimated system state, the optimized nonlinear system model and the predefined control target; constructing the system state prediction equation according to the nonlinear system model; defining the cost function related to the control target, including the state deviation term, the control input term and the smoothing term; setting the constraints related to the system physical limitations and performance requirements; combining the cost function and the constraints, constructing the model predictive control optimization problem, and obtaining the model predictive optimization control framework; Step S32, obtaining the estimated system state, the optimized nonlinear system model and the current execution efficiency data; applying the dynamic time domain adjustment algorithm to calculate the system characteristic time constant according to the system model; And evaluate the computing resource load in combination with the current execution efficiency; dynamically adjust the prediction time domain length according to the characteristic time constant and computing resource load to ensure that the prediction covers the dynamics of key systems; adjust the control time domain length to balance the control performance and computational complexity; obtain the optimized prediction time domain and control time domain through an iterative optimization process; Step S33, obtaining an estimated system state, a covariance matrix and predefined safety margin parameters; Use the probabilistic constraint generation algorithm to build a state distribution model based on the mean and covariance of the state estimate; determine the allowable probability of constraint violation based on the safety margin parameter; use the chance constraint theory to transform deterministic constraints into probabilistic constraints; for nonlinear constraints, use linearization or sampling methods to perform probabilistic processing; evaluate the feasibility of probabilistic constraints through numerical integration or Monte Carlo methods, and finally obtain a robust control constraint set; Step S34, obtaining a model prediction optimization control framework, an optimized prediction time domain and control time domain, and a robust control constraint set; constructing a multi-objective optimization problem, decomposing the original cost function into multiple sub-objectives, including control performance, energy efficiency, and constraint satisfaction; then introducing weight coefficients to form a weighted and multi-objective function; integrating the robust control constraint set into the optimization problem so that it satisfies the constraint conditions under state uncertainty; adjusting the dimension of the optimization problem according to the optimized prediction and control time domains; combining the multi-objective function, constraint conditions, and optimization variable range to obtain a complete optimization problem model; The step S4 is specifically as follows: Step S41, obtaining a complete optimization problem model and information on currently available computing resources of the system; applying an adaptive solver selection algorithm to analyze the characteristics of the optimization problem, including problem size, constraint type, and properties of the objective function; evaluating currently available computing resources, including processor performance, memory capacity, and parallel computing capabilities; selecting the most suitable optimization algorithm from a predefined solver library based on problem characteristics and resource conditions; configuring parameters of the selected solver, including convergence threshold, maximum number of iterations, and numerical precision, to obtain a configured solver; Step S42, obtaining a complete optimization problem model and a configured solver; using a parallel computing framework, decomposing the optimization problem into multiple sub-problems; and solving these sub-problems simultaneously on multiple processing cores; In the solution process, computing resources are dynamically allocated to balance the computing load of each sub-problem; through iterative calculation and information exchange, the global optimal solution is gradually converged; the optimal control sequence is obtained, which includes the hopper speed v(t) and displacement d(t) control parameters for multiple future time steps; Step S43, obtaining the optimal control sequence, the optimized nonlinear system model and the predefined uncertainty description; performing Monte Carlo simulation, and generating a large number of random disturbance samples according to the uncertainty description; For each sample, apply the optimal control sequence to the system model, simulate the system response and collect all simulation results, calculate statistical indicators, including average performance, worst-case performance and performance variance; calculate the robustness index of the control strategy based on the statistical indicators; obtain sensitivity analysis results by analyzing the impact of different uncertain factors on system performance; Step S44, obtain the current control parameter u*(t) = {v(t), d(t)} in the optimal control sequence and the current control cycle parameter Tc of the system; execute the control action, and send the speed v(t) and displacement d(t) instructions to the hopper drive system; within the control cycle Tc, continuously monitor the actual motion state of the hopper and the system response; collect real-time sensor data, including the actual position, speed and acceleration information of the hopper; process the collected data, calculate the accuracy and response time of the control execution, and obtain the execution efficiency E(t); compare the actual executed hopper motion parameters with the control instructions, and obtain the actual executed hopper control parameters u_actual(t) = {v_actual(t), d_actual(t)}; integrate all monitoring and calculation results to obtain the complete actual system response Y(t).

2. The single module follow-up control method according to claim 1, It is characterized in that The step S1 is specifically as follows: Step S11, obtaining an original image sequence captured by a high-speed industrial camera; applying an adaptive exposure algorithm to the original image sequence, dynamically adjusting exposure parameters according to the overall brightness distribution of the image, until the contrast and detail visibility of the image reach a preset value, thereby obtaining an optimized image sequence; Step S12, obtaining the original vibration signal collected by the vibration sensor; applying wavelet transform to the original vibration signal, decomposing the signal into different frequency components, identifying and removing high-frequency noise components, and then reconstructing the signal to obtain a vibration signal after noise reduction; Step S13, obtaining the hopper position data recorded by the encoder; applying the Kalman filter algorithm to the hopper position data, combining the system dynamic model and the measurement data, estimating the real position state, filtering out random errors, and obtaining the smoothed hopper position data; Step S14, obtaining the optimized image sequence, the vibration signal after noise reduction and the smoothed hopper position data; A timestamp alignment algorithm is applied to align the data of different modalities onto a unified time axis according to the acquisition timestamps of each data, thus obtaining a synchronized multimodal data set. Step S15, obtaining a synchronized multimodal data set; applying an improved SURF algorithm to the image sequence in the multimodal data set, extracting key points and descriptors in the image, and obtaining tablet features; Apply fast Fourier transform to the vibration signal to convert the time domain signal into frequency domain representation and obtain frequency domain features; Apply the difference algorithm to the hopper position data, calculate the position difference and second-order difference of adjacent time points, and obtain the velocity and acceleration features; combine the tablet features, frequency domain features, velocity and acceleration features to obtain the feature set; Step S16, obtaining a feature set and a predefined basic parameter set; applying an adaptive parameter adjustment algorithm to the feature set, calculating and generating a set of initialized adjustable parameter values ​​according to the statistical characteristics of the features and the predefined basic parameters, and obtaining an initialized adjustable parameter set.

3. The single module follow-up control method according to claim 1, It is characterized in that The step S2 is specifically as follows: Step S21, obtaining a feature set and an initialized adjustable parameter set; applying recursive least squares method to the feature set and the adjustable parameter set, updating the model parameter estimation, using a forgetting factor to give a higher weight to recent data, and identifying the dynamic characteristics of the system; by minimizing the weighted prediction error sum of squares, iteratively calculating the model parameters, and obtaining the nonlinear system model parameters; Step S22, obtaining nonlinear system model parameters and a predefined set of candidate model structures; applying the Bayesian model selection algorithm to each candidate model structure, calculating the marginal likelihood of the model and combining it with the prior probability to obtain the posterior probability of each model; comparing the posterior probabilities of all candidate model structures, selecting the candidate model structure with the highest posterior probability, and obtaining an optimized nonlinear system model; Step S23, obtaining a feature set and an optimized nonlinear system model; applying a delayed coordinate embedding method to calculate the mutual information function and determine the optimal time delay; and using a pseudo nearest neighbor algorithm to determine a suitable embedding dimension; performing a delayed coordinate transformation on the feature data according to the determined time delay and embedding dimension, reconstructing the system state space, and obtaining a reconstructed state vector; Step S24, obtaining a reconstructed state vector and a synchronized multimodal data set; applying an adaptive fault-tolerant Kalman filter algorithm, first predicting the state according to the system model; then updating the state using multi-source measurement data; in this process, dynamically adjusting the process noise covariance and the measurement noise covariance to adapt to changes in the system state and measurement characteristics; Possible sensor failures or abnormal measurements are identified and processed by introducing a fault detection mechanism; through iterative prediction and update steps, the estimated system state and the corresponding covariance matrix are obtained.

4. The single module follow-up control method according to claim 1, It is characterized in that Step S12 is specifically as follows: Step S121, obtaining an original vibration signal collected by a vibration sensor; applying an adaptive segmentation algorithm to the original vibration signal, dividing the signal into at least two time periods according to the local statistical characteristics of the signal, and obtaining a segmented vibration signal sequence; Step S122, obtaining the segmented vibration signal sequence, applying an adaptive basis function selection algorithm to each signal segment, selecting the most suitable wavelet basis function from a predefined basis function library according to the frequency characteristics of the signal, and obtaining the optimal wavelet basis function corresponding to each signal segment; Step S123, obtaining the segmented vibration signal sequence and the optimal wavelet basis function; applying the corresponding optimal wavelet basis function to each signal segment, performing multi-scale wavelet decomposition, and obtaining the wavelet coefficient of each signal segment; Step S124, obtaining wavelet coefficients, and applying an adaptive threshold algorithm, calculating an optimal threshold according to the statistical distribution of the wavelet coefficients at each scale, performing soft threshold processing on the wavelet coefficients, and obtaining the denoised wavelet coefficients; Step S125, obtaining the denoised wavelet coefficients and the optimal wavelet basis function, using inverse wavelet transform to reconstruct each signal segment, and then splicing all the reconstructed signal segments to obtain a complete denoised vibration signal; Step S126, obtaining the vibration signal after noise reduction, applying an adaptive smoothing algorithm, dynamically adjusting the smoothing window size according to the local curvature characteristics of the signal, performing final smoothing processing on the signal, and obtaining an optimized noise reduction vibration signal.

5. The single module follow-up control method according to claim 1, It is characterized in that Step S15 is specifically as follows: Step S151, obtaining an optimized image sequence in a synchronous multimodal data set, and applying an improved SURF algorithm to construct a scale space pyramid, and then using a Hessian matrix to detect feature points at each scale space layer to obtain a set of candidate feature points; Step S152, obtaining a set of candidate feature points, applying an adaptive non-maximum suppression algorithm, dynamically adjusting the suppression window size according to the complexity of the local image structure, screening the most significant feature points, and obtaining an optimized feature point set; Step S153, obtaining an optimized feature point set, constructing a rotation invariant descriptor for each feature point, calculating the main direction of the feature point neighborhood, and calculating the Haar wavelet response in the main direction coordinate system to obtain a feature point descriptor set; Step S154, obtaining the denoised vibration signal in the synchronous multimodal data set, applying an adaptive time-frequency analysis algorithm, dynamically adjusting the time and frequency resolution according to the non-stationary characteristics of the signal, and obtaining a time-frequency representation of the signal; Step S155, obtaining the obtained time-frequency representation, using a peak detection algorithm to identify significant peaks on the time-frequency plane, extracting the time, frequency and amplitude information of the significant peaks, and obtaining a feature vector of the vibration signal; Step S156, obtaining the smoothed hopper position data in the synchronous multimodal data set, using an adaptive polynomial fitting algorithm, dynamically selecting the most suitable polynomial order according to the local complexity of the data, performing piecewise fitting on the position data, and obtaining a polynomial representation of the position; Step S157, obtaining a position polynomial representation, calculating the first and second order derivatives of the polynomial, and obtaining velocity and acceleration estimates; applying a statistical feature extraction algorithm to the velocity and acceleration sequence, calculating statistics including mean, variance, and peak value, and obtaining a motion feature vector; Step S158, obtain a feature point descriptor set, a vibration feature vector, and a motion feature vector, perform normalization processing, and use the weighted summation method in the feature fusion algorithm to combine the features of different modes, wherein the weight is dynamically adjusted according to the discriminative ability of each feature to obtain a fused multi-modal feature set.

6. The single module follow-up control method according to claim 1, It is characterized in that Step S23 is specifically as follows: Step S231, obtaining a feature set and an optimized nonlinear system model, applying a mutual information algorithm to calculate the mutual information between the time series of each dimensional data in the feature set and its different time delayed versions, and obtaining a mutual information function curve; Step S232, obtaining a mutual information function curve, applying an adaptive threshold selection algorithm to identify the first local minimum of the mutual information function, determining an optimal time delay τ, and obtaining a reconstructed time delay parameter; Step S233, obtaining a feature set, an optimized nonlinear system model and an optimal time delay τ, applying an improved false nearest neighbor algorithm, sequentially increasing the embedding dimension, calculating the false nearest neighbor ratio under each dimension, until the false nearest neighbor ratio drops below a preset threshold, and obtaining the optimal embedding dimension m; Step S234, obtaining a feature set, an optimal time delay τ and an optimal embedding dimension m, and constructing a delay coordinate matrix for each time series in the feature set, wherein each row of the matrix represents a reconstructed state vector, to obtain a preliminary reconstructed state space; Step S235, obtaining a preliminary reconstructed state space and an optimized nonlinear system model; applying a principal component analysis algorithm to calculate the covariance matrix of the reconstructed state space, performing eigenvalue decomposition, selecting the first k principal components with the largest contribution rate, and obtaining a reduced-dimensional state space representation; k is a natural number; Step S236, obtain the state space representation after dimensionality reduction, apply the local linear embedding algorithm to find the nearest neighbor point for each state point, construct local reconstruction weights, and then use these weights to reconstruct the data points in the low-dimensional space to obtain the final optimized reconstructed state vector.

7. The single module follow-up control method according to claim 1, It is characterized in that Step S24 is specifically as follows: Step S241, obtaining the reconstructed state vector and the synchronized multimodal data set, applying an adaptive noise estimation algorithm, utilizing the statistical characteristics of the innovation sequence, dynamically estimating the process noise covariance Q(k) and the measurement noise covariance R(k), and obtaining an adaptive noise model; Step S242, obtaining an adaptive noise model, reconstructing a state vector and a synchronous multimodal data set, applying a strong tracking filtering algorithm, introducing an attenuation factor, dynamically adjusting the prediction error covariance matrix, enhancing the filter's adaptability to model mismatch, and obtaining a corrected prediction error covariance; Step S243, obtaining the corrected prediction error covariance, the reconstructed state vector and the synchronized multimodal data set, applying the adaptive factor method, calculating the adaptive factor according to the statistical characteristics of the new information sequence, weighting the contribution of the measurement update, and obtaining the adaptive gain matrix; Step S244, obtaining an adaptive gain matrix, a reconstructed state vector and a synchronized multimodal data set, executing a prediction step of a Kalman filter, using a system model to make a one-step prediction of the state, and obtaining a priori state estimation and error covariance; Step S245, obtaining a priori state estimation and error covariance, an adaptive gain matrix and a synchronous multimodal data set; executing a Kalman filter update step, correcting the a priori estimation in combination with the measurement data, and obtaining a posteriori state estimation and error covariance; Step S246, obtaining a posteriori state estimation and error covariance, applying a fault detection and isolation algorithm to calculate a normalized innovation sequence, and performing hypothesis testing to identify possible sensor faults or abnormal measurements to obtain a fault diagnosis result; Step S247, obtaining a posteriori state estimation, error covariance, and fault diagnosis results; The robust data fusion algorithm is applied to adjust the credibility of each sensor data according to the fault diagnosis result, and the fused state estimation and uncertainty are recalculated to obtain the final estimated system state and covariance matrix.

8. A single module follow-up control system, It is characterized in that include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the single-module follow-up control method described in any one of claims 1 to 7.