Crane closed-loop control method, system and computer readable storage medium

Abstract

The application discloses a crane closed-loop control method, system and computer readable storage medium, and belongs to the technical field of hoisting machinery. The closed-loop control method comprises the following steps: constructing a multi-frequency adaptive disturbance observer, decomposing a composite disturbance vector into a low-frequency component vector, a medium-frequency component vector and a high-frequency component vector, wherein an extended state observer is used to estimate the low-frequency component vector, an adaptive radial basis function neural network is used to estimate the medium-frequency component vector, and a high-order sliding mode observer based on a super-spiral algorithm is used to estimate the high-frequency component vector; a low-frequency disturbance estimation value, a medium-frequency disturbance estimation value and a high-frequency disturbance estimation value are combined into a total disturbance estimation value, and the total disturbance estimation value is output as a feedforward compensation to a crane controller for control, so that the full frequency domain range can be covered, the suppression ratio of the composite disturbance can be effectively improved, and online tracking of time-varying parameters can be realized.

Description

Technical Field

[0001] This invention relates to the field of crane machinery technology, and in particular to a closed-loop control method, system and computer-readable storage medium for cranes. Background Technology

[0002] Cranes, as typical underactuated systems, have fewer drive inputs than system degrees of freedom, and exhibit a strong coupled nonlinear relationship between load oscillation and mechanism motion. They are widely used for loading and unloading goods in docks, factories, construction sites, and other locations. With the improvement of production technology, cranes are developing towards intelligence, larger size, and faster speed, placing higher demands on system response speed and control precision.

[0003] Existing crane control technologies are mainly divided into two categories: open-loop control and closed-loop control.

[0004] Open-loop control techniques mainly include input shaping methods (such as ZV, ZVD, EI, etc.), which eliminate oscillations at specific frequencies by convolving pulse sequences with the original commands. These methods do not require angle sensors and have low implementation costs, but their performance degrades significantly when faced with external disturbances and parameter changes, and they cannot handle modeling errors and time-varying disturbances.

[0005] Closed-loop control technology has developed rapidly in recent years, mainly including the following technical routes: PID control and its improved version: The IPID-ADOB control strategy employs an adaptive disturbance observer as a feedforward compensator to estimate and compensate for complex disturbances in real time. This method optimizes controller parameters using the dung beetle algorithm, with ITSE as the objective function, and achieves good results in simulations. Sliding mode control: While employing discontinuous control to achieve robust control, it suffers from chattering, which can negatively impact the lifespan of the actuator. Building upon this, some researchers have proposed an improved sliding mode control based on the sigmoid function, using smoothing techniques to reduce chattering. Adaptive fuzzy control: For tower cranes with variable rope length, fuzzy logic is used to approximate system uncertainties, and safety control is achieved by combining state constraints.

[0006] However, most existing technologies simplify complex disturbances into a single concentrated disturbance term using a single ESO (Electronic Optimizer). The observation bandwidth of the ESO is limited by the system sampling frequency and noise characteristics, making it impossible to simultaneously ensure the observation accuracy of low, medium, and high frequency disturbances. When the frequency domain components of the disturbance exceed the observer bandwidth, the observation accuracy decreases significantly; conversely, when the observer bandwidth is increased to capture high-frequency disturbances, the measurement noise is amplified, leading to high-frequency oscillations in the estimated values. This contradiction severely restricts the robustness of existing methods under full-band disturbance environments.

[0007] Furthermore, current applications of digital twin technology in the crane industry are mainly concentrated in offline or semi-offline scenarios such as collision prediction, structural optimization, and operation monitoring, and have not yet been deeply integrated with real-time control loops. The predictive capabilities of digital twin models have not been effectively used for disturbance compensation and online optimization of controller parameters, and cannot provide real-time correction information to the controller within millisecond-level control cycles. This technological gap also limits the potential for crane control systems to develop towards higher levels of intelligence. Summary of the Invention

[0008] This application provides a closed-loop control method, system, and computer-readable storage medium for a crane, which can cover the full frequency range of 0-50Hz, effectively improve the suppression ratio of composite disturbances, and can also achieve online tracking of time-varying parameters through adaptive correction driven by twin bias.

[0009] The first aspect of this application provides a closed-loop control method for a crane, comprising the following steps: Construct a multi-band adaptive disturbance observer to decompose the composite disturbance vector d into low-frequency component vector d l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h ,in: The low-frequency component vector is estimated using an extended state observer, and the state estimation equation is as follows: =A e z+B eu +L l (y C e z), where, To extend the first derivative of the state estimate z with respect to time, A e To expand the system matrix, z=[x` T , d` l T ] T Where x' is the system state estimate, d' l B is the estimated value for low-frequency disturbance. e To expand the input matrix, u L is the driving force of the motor. l Let y be the gain matrix of the low-frequency observer, and y be the output vector of the actual system. e To expand the output matrix, the low-frequency disturbance estimate d` l The update pattern is ` l = ωl d`l + ωl L l (y y`), where, ` l The low-frequency disturbance estimate d` l The first derivative with respect to time, ωl Here, represents the bandwidth of the low-frequency observer, in rad / s; y is the actual system output vector; y' is the estimated system output value; and y' = C. e z; An adaptive radial basis function neural network is used to estimate the intermediate frequency component vector. The output of the neural network is: d` m =W T Φ(x), where d` m W is the estimated value of the intermediate frequency disturbance. T Let W be the transpose of the network weight matrix W, where W has dimension 1. N ×2, where N Let denoted as the number of neurons, and Φ(x) be the Gaussian kernel vector with dimension . N ×1, its first i The components are: i (x) ,in i (x) is the first i The output of each neuron It is a natural exponential function. Let the system state vector x and the first i vector of Gaussian kernel center point The difference in Euclidean distance when laid flat, For the first i The width of a Gaussian kernel, where the adaptive update rule of the weight matrix W is: =ΓΦ(x) K m W, where, Let W be the first derivative of the weight matrix W with respect to time, and Γ be the adaptive gain matrix. N × N A positive definite diagonal matrix, Let e ​​be the observation error vector m =y The transpose of y', K m for The coefficient matrix of the correction term is N × N matrix; A high-order sliding mode observer based on the superspiral algorithm is used to estimate the high-frequency component vector, which has the following continuous form: h = λ 1 sgn(e)+ λ 2∫sgn(e) dt In the formula, h The high-frequency disturbance estimate d` h The first derivative with respect to time, λ 1 represents the sliding mode gain, used to control the convergence speed. Let the observation error vector be e=y Let y' be a vector formed by the square roots of the absolute values ​​of each component, and sgn(e) be the coincidence function of the observation error vector, where each component takes the value +1, -1, or 0. λ 2 represents the integral gain, ∫sgn(e) dt The discretized form of the high-frequency observer is the integral of the sign function vector over time: d` h ( k +1)=d` h ( k )+ Ts [ λ 1 sgn(e( k ))+ λ 2v( k )],v( k +1)=v( k )+ Ts sgn(e( k In the formula, d` h ( k +1) indicates the first k +1 sampling time point high-frequency disturbance estimate, d` h ( k ) is the first k The high-frequency disturbance estimate at each sampling time. Ts The sampling period is λ 1 represents the sliding mode gain. For the first The square root vector of the absolute value of the observation error at each time point, sgn(e( k )) is the first k The sign function vector of the observation error at each sampling time. λ 2 represents the integral gain, v( k ) is the first k The integral state vector at each sampling time point, v( k +1) is the first k The integral state vector at +1 sampling time points; The low-frequency disturbance estimate d` lMid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

[0010] In one possible implementation, the low-frequency component has a frequency range of 0-0.5Hz, the mid-frequency component has a frequency range of 0.5-5Hz, and the high-frequency component has a frequency range of 5-50Hz.

[0011] In one possible implementation, the low-frequency disturbance estimate d` l Through a low-pass filter Gl ( s )= ωl / ( s + ωl Extract, where ωl The bandwidth of the low-frequency observer is... s The Laplace operator; the estimated mid-frequency disturbance d` m Through bandpass filter Gm ( s )= Extract, of which ξm For the damping ratio, ωm The center frequency; the estimated high-frequency disturbance value d` h Extracted using a high-pass filter.

[0012] In one possible implementation, based on the past T Energy output by each observer within seconds Pi = dτ The composite weights are dynamically adjusted, and the estimated total composite perturbation is d`= αl d` l + αm d` m + αh d` h The weighting coefficient αi = , i = l , m , h ,∥d` i ∥ represents the Euclidean norm of the estimated value.

[0013] In one possible implementation, the crane closed-loop control method further includes constructing a digital twin high-fidelity dynamic model: M(q)q``+C(q,q`)q`+G(q)+F fric =u+d ext In the formula, M(q) is the mass matrix dependent on the generalized coordinate vector q, q= ,in For the displacement of the trolley, For the load swing angle, The length of the rope. Let q'' be the elastic deformation of the wire rope, q'' be the generalized acceleration vector, C(q,q') be the Coriolis force and centrifugal force matrix, q' be the generalized velocity vector, G(q) be the gravity term vector, and F be the force vector. fric Let u be the friction force vector, and d be the control force vector. ext The external disturbance vector; Define the twin bias vector: e twin =y physical y virtual In the formula, y physical y is the actual output vector of the physical system. virtual The output vector of the digital twin virtual model; Establish a deviation prediction model: e` twin =A twin e twin +B twin Δp, where e` twin For twin bias vector e twin The first derivative with respect to time, A twin Let B be the deviation state matrix. twin The deviation input matrix is ​​Δp, which is the unmodeled parameter deviation vector, including the change in friction coefficient and the degradation of wire rope stiffness. Online identification of time-varying parameter vectors using the recursive least squares method: p( t )= ,in For load quality, For friction, Here is the stiffness coefficient of the wire rope; the parameter update rule is: P` ( k +1)=p`( k )+K RLS ( k )[ y ( k ) ( k )p`( k In the formula, p` ( k +1) is the firstk +1 sampling time point parameter estimation vector, p`( k ) is the first k The parameter estimation vector at each sampling time point, K RLS ( k ) is the first k The recursive least-squares gain vector at each sampling time. y ( k ) is the first k The system output scalar at each sampling time. ( k ) is the first k Regression vector at each sampling time ( k The transpose of ) The update rule for the gain matrix is: K RLS ( k )= In the formula, p ( k -1) is the first k The covariance matrix at -1 sampling time point, λ This is the forgetting factor, with a value range of 0-1; The update rule for the covariance matrix is: P( k )= [P( k 1) K RLS ( k ) ( k )P( k 1)], where P( k ) is the first k The covariance matrix at each sampling time point is used to perform fusion correction on the intermediate frequency observer and the high frequency observer using twin bias: =d m +K twin e twin , =d h +K high e` twin In the formula, The corrected intermediate frequency disturbance estimate, d m K is the original intermediate frequency disturbance estimate. twin This is the twin bias correction gain matrix. The corrected high-frequency disturbance estimate, d h K is the original high-frequency disturbance estimate. high This is the high-frequency correction gain matrix; The corrected mid-frequency disturbance estimate and high-frequency disturbance estimates Replace the original estimate with the low-frequency disturbance estimate d` l The values ​​are combined to form a total disturbance estimate.

[0014] In one possible implementation, the forgetting factor of the recursive least squares method λ The value is 0.98, and the initial covariance matrix P(0) = I3×3, where I3×3 is a 3x3 identity matrix, and the sampling interval is... Ts =0.01 seconds.

[0015] In one possible implementation, the elastic deformation of the wire rope in the digital twin high-fidelity dynamic model... δ The dynamics are based on the stiffness coefficient kl and damping coefficient cl Confirmed, satisfied mlδ ``+ clδ `+ klδ = T ,in ml For load quality, δ `` represents elastic acceleration cl The damping coefficient is... δ `is the elastic deformation velocity` kl This is the stiffness coefficient. T This refers to the tension of the steel wire rope.

[0016] In one possible implementation, a micro-signal analysis unit is integrated into the digital twin server to continuously monitor the physical signal characteristics of the optical fiber transmission link. Based on the physical signal characteristics, a transmission channel status score is established, and a trend prediction algorithm is used to predict the future health status of the optical fiber medium. The predicted future status score is continuously compared with a preset warning line, and a graded response is triggered. In the case of a mild warning, the disturbance observer correction gain is automatically adjusted to reduce the dependence on digital twin correction information. In the case of a severe warning, a preventive maintenance alarm is issued.

[0017] A second aspect of this application provides a closed-loop control system for a crane, used to implement the closed-loop control method for a crane as described above. The control system includes: The multi-band adaptive disturbance observer building module is used to construct a multi-band adaptive disturbance observer, which decomposes the composite disturbance vector d into low-frequency component vectors di. l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h ; The low-frequency component vector estimation submodule is used to estimate the low-frequency component vector using an extended state observer; The intermediate frequency component vector estimation submodule is used to estimate the intermediate frequency component vector using an adaptive radial basis function neural network; The high-frequency component vector estimation submodule is used to estimate the high-frequency component vector using a high-order sliding mode observer based on the superspiral algorithm; The fusion output module is used to convert the low-frequency disturbance estimate d` l Mid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

[0018] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the crane closed-loop control method as described above.

[0019] Beneficial effects: Compared with the prior art, the crane closed-loop control method, system and computer-readable storage medium provided in this application decompose the complex disturbance into low, medium and high frequency bands and design observers for each band, solving the problem that the existing single observer cannot cover the entire frequency band. Among them, the low frequency adopts ESO, the medium frequency adopts adaptive RBF network and the high frequency adopts high-order sliding mode to form a complete disturbance suppression spectrum, which enables the crane positioning accuracy to reach ±2mm, significantly improves the ability to suppress periodic disturbances such as wind load and elastic vibration, and can improve the response speed to rapid disturbances such as gear meshing and track impact, improving the transient response speed by about 40%. Meanwhile, it can also form closed-loop control through adaptive correction driven by digital twin bias, which can improve parameter perturbation tolerance by 66% and improve time-varying parameter tracking delay by more than 60%, transforming the observer from "passive estimation" to "active cognition", providing a new technical path for intelligent closed-loop control of cranes; It can also shift the monitoring focus from data to the microscopic physical state of the data channel. By predicting fiber degradation, it can trigger preventive maintenance or system adaptive calibration before problems occur, avoiding the recursive least squares algorithm being exposed to severely polluted data streams for a long time. This fundamentally ensures its robustness and the accuracy of the digital twin model, and solves the hidden, long-term accumulated system health risks.

[0020] These and other objects, features and advantages of the present invention will become fully apparent from the following detailed description. Attached Figure Description

[0021] Figure 1 A schematic flowchart of the crane closed-loop control method of this application is shown. Detailed Implementation

[0022] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0023] Those skilled in the art should understand that, in the disclosure of this specification, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting the present invention.

[0024] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will also be understood that terms, such as those defined in commonly used dictionaries, shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art and shall not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0025] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0026] Cranes experience complex disturbances with multi-frequency characteristics during actual operation. Low-frequency disturbances mainly originate from modeling errors and parameter perturbations (such as changes in load mass and friction coefficient drift); mid-frequency disturbances arise from wind loads, elastic vibrations of wire ropes, and changes in load inertial forces; and high-frequency disturbances stem from gear meshing impacts, track irregularities, and sensor measurement noise. The dynamic characteristics of disturbances in different frequency bands differ significantly, requiring different observation and compensation strategies.

[0027] Current applications of digital twin technology in the crane industry are mainly concentrated in offline or semi-offline scenarios such as collision prediction, structural optimization, and operation monitoring, and have not yet been deeply integrated with real-time control loops. The predictive capabilities of digital twin models have not been effectively used for disturbance compensation and online optimization of controller parameters, and cannot provide real-time correction information to the controller within millisecond-level control cycles. This technological gap limits the potential for crane control systems to develop towards higher levels of intelligence.

[0028] refer to Figure 1 The first aspect of this application provides a closed-loop control method for a crane, comprising the following steps: First, a multi-band adaptive disturbance observer is constructed.

[0029] Perturbation decomposition: Decompose the composite perturbation vector d into low-frequency component vector d l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h This frequency band decomposition is the core technical concept of this scheme, and its necessity and uniqueness are mainly reflected in the following aspects: Physical Differences in Disturbance Characteristics: Crane systems face diverse sources of complex disturbances during actual operation, with their physical characteristics varying significantly across different frequency bands. For example, load mass changes and model parameter perturbations are slowly changing or quasi-static disturbances, primarily concentrated in the low-frequency band; while wind loads and wire rope elastic vibrations exhibit periodic or narrow-band random disturbances, mainly concentrated in the mid-frequency band; gear meshing impacts and rail joint impacts are transient, high-frequency disturbances. The dynamic mechanisms, amplitudes, and frequency characteristics of these disturbances are completely different.

[0030] The inherent limitations of a single observer: Any single observer algorithm has its optimal applicability and inherent performance bottlenecks. For example, the Extended State Observer (ESO) is very effective in handling low-frequency, slowly varying disturbances, but its bandwidth is limited, making it difficult to track high-frequency dynamics; neural networks excel at learning and approximating nonlinear, periodic disturbances in the mid-frequency band, but their response to sudden high-frequency impacts is insufficient; sliding mode observers respond quickly to high-frequency disturbances and are robust, but their discontinuous switching characteristics introduce chattering when used in the low-frequency band. Therefore, attempting to use a single observer to handle disturbances in all frequency bands will result in poor performance in some frequency bands, failing to achieve disturbance suppression across the entire frequency band, which is a fundamental technical contradiction.

[0031] The Uniqueness and Necessity of the "Divide and Conquer" Approach: This solution is the first to systematically apply the "divide and conquer" approach to crane control. Instead of simply combining several observers, it first performs spectral analysis and decoupling of the complex disturbance, then "tailor-makes" the most suitable observer for each decomposed frequency band. The uniqueness of this strategy lies in its decomposition of a complex and difficult-to-handle full-band disturbance suppression problem into three relatively simple and well-defined sub-problems, which are then solved one by one using optimized tools. This is a completely new technical path in crane control, and its necessity lies in the fact that only in this way can the performance ceiling of a single observer be broken through, achieving high-precision, high-dynamic collaborative suppression of disturbances across the 0-50Hz full-band—an essential path to achieving ultra-high-precision crane control.

[0032] Specifically: The low-frequency component vector is estimated using an extended state observer, and the state estimation equation is as follows: =A e z+B eu +L l (y C e z), where, To extend the first derivative of the state estimate z with respect to time, A e To expand the system matrix, z=[x` T , d` l T ] T Where x' is the system state estimate, d' l B is the estimated value for low-frequency disturbance. e To expand the input matrix, u L is the driving force of the motor. l Let y be the gain matrix of the low-frequency observer, and y be the output vector of the actual system. e To expand the output matrix, the low-frequency disturbance estimate d` l The update pattern is as follows: ` l = ωl d` l + ωl L l (y y`), where, ` l The low-frequency disturbance estimate d` l The first derivative with respect to time, ωl The low-frequency observer bandwidth is expressed in rad / s, for example, 5 rad / s. y represents the actual system output vector, y' represents the estimated system output, and y' = C.e z; The gain matrix of the low-frequency observer is designed as L using the pole placement method. l =[15, 75, 125] T The observer can estimate disturbance components in the frequency range of 0-0.8 Hz, with a convergence time of approximately 0.8 s.

[0033] An adaptive radial basis function neural network (RBF neural network) is used to estimate the intermediate frequency component vector. The output of the neural network is: d` m =W T Φ(x), where d` m W is the estimated value of the intermediate frequency disturbance. T Let W be the transpose of the network weight matrix W, where W has dimension 1. N ×2, where N For example, the number of neurons. N =20, Φ(x) is the Gaussian kernel function vector with dimension . N ×1, that is 20 The output vector of the Gaussian kernel of ×1, its first... i The components are: i (x) , i =1,2,…,20, where i (x) is the first i The output of each neuron It is a natural exponential function. Let the system state vector x and the first i vector of Gaussian kernel center point The difference in Euclidean distance when laid flat, For the first i The width of a Gaussian kernel, for example =0.3, where the adaptive update rule of the weight matrix W is: =ΓΦ(x) K m W, where, Let W be the first derivative of the weight matrix W with respect to time, and Γ be the adaptive gain matrix. N × N A positive definite diagonal matrix, Γ=0.01 I20×20 (a 20th order identity matrix multiplied by 0.01). Let e ​​be the observation error vector m =y The transpose of y', K m for The coefficient matrix of the correction term isN × N Matrix, σ, correction term coefficient matrix K m =0.001 I20×20; This observer can approximate mid-frequency nonlinear disturbances in real time and has a significant effect on suppressing periodic disturbances such as wind load fluctuations.

[0034] A high-order sliding mode observer based on the superspiral algorithm is used to estimate the high-frequency component vector, which has the following continuous form: h = λ 1 sgn(e)+ λ 2∫sgn(e) dt In the formula, h The high-frequency disturbance estimate d` h The first derivative with respect to time, λ 1 represents the sliding mode gain, used to control the convergence speed. Let the observation error vector be e=y y' is a vector formed by the square roots of the absolute values ​​of each component, and sgn(e) is the coincidence function of the observation error vector, with each component taking the value of +1 (positive), -1 (negative), or 0 (zero). λ 2 represents the integral gain, ∫sgn(e) dt The discretized form of the high-frequency observer is the integral of the sign function vector over time: d` h ( k +1)=d` h ( k )+ Ts [ λ 1 sgn(e( k ))+ λ 2v( k )],v( k +1)=v( k )+ Ts sgn(e( k In the formula, d` h ( k +1) indicates the first k +1 sampling time point high-frequency disturbance estimate, d` h ( k ) is the first k The high-frequency disturbance estimate at each sampling time. Ts For example, the sampling period Ts =0.01s, λ 1 represents the sliding mode gain, for example... λ 1 = 5 For the first The square root vector of the absolute value of the observation error at each time point, sgn(e( k )) is the first k The sign function vector of the observation error at each sampling time. λ 2 represents the integral gain, for example... λ 2=3, v( k ) is the first k The integral state vector at each sampling time point, v( k +1) is the first k The observer can estimate high-frequency disturbances such as gear impact and track irregularities within 0.1s by integrating the state vector at +1 sampling time.

[0035] The low-frequency disturbance estimate d` l Mid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

[0036] In the crane closed-loop control method provided in this application, an extended state observer (ESO) is used to estimate the low-frequency component dl, which has a good effect on handling low-frequency slow-varying disturbances. An adaptive radial basis function (RBF) neural network is used to estimate the mid-frequency component dm, which is good at learning and approximating nonlinear and periodic disturbances in the mid-frequency range. A high-order sliding mode observer based on the superspiral algorithm is used to estimate the high-frequency component dh, which can respond quickly to high-frequency disturbances and has strong robustness. By integrating the extended state observer (ESO), neural network and sliding mode control, which have very different principles, with the digital twin model, the following key technical challenges are mainly solved: The challenge of synergy and consistency: The three observers are based on different mathematical principles, and their output perturbation estimates differ in dynamic response, phase delay, and noise characteristics. Simply adding them together can lead to mismatches and conflicts, and may even cause system oscillations. This scheme introduces specific filter banks (e.g., low-pass, band-pass, and high-pass) to "shape" the output of each observer, ensuring that each estimate mainly functions within its designated frequency band. This achieves orthogonal decomposition in the frequency domain, guaranteeing the smoothness and consistency of the fusion. Seamless full-band coverage and handover challenges: Achieving a smooth and seamless transition when the disturbance frequency crosses different frequency bands is a technical challenge. For example, when a disturbance changes from 4Hz to 6Hz, the dominant observer needs to smoothly transition from a neural network to a sliding mode observer. This solution addresses this challenge by using a dynamic weighted fusion method (described later) for energy weighting, rather than a hard handover. The weights are dynamically adjusted based on the output energy of each observer to ensure a smooth transition at frequency band boundaries.

[0037] In some embodiments, based on in-depth analysis of the dynamic characteristics of the crane system and a large amount of experimental data and simulation verification, the frequency band of the crane is divided as follows: The low-frequency component has a frequency range of 0-0.5Hz, which mainly corresponds to the slow time-varying uncertainty in the system model, such as the slow change of load mass during hoisting, the slow drift of friction coefficient caused by mechanical structure aging or temperature change, etc. These disturbances are close to DC or change very slowly, and belong to quasi-static disturbances. Their spectral energy is mainly concentrated below 0.5Hz. The frequency range of the mid-frequency component is 0.5-5Hz. It mainly captures the main vibration modes of the system and common external environmental disturbances. For example, the natural swaying frequency of the crane load under typical working conditions and the elastic vibration frequency of the wire rope usually fall within this range. At the same time, the main energy spectrum of the structural swaying caused by natural wind load is also concentrated in this frequency band. The frequency range of the high-frequency component is 5-50Hz, which corresponds to the rapid and transient impacts and high-frequency noise in the system. This mainly includes the impact generated when the gears mesh in the transmission system, the impact vibration generated when the trolley rolls over the track joints or uneven areas, and the high-frequency noise introduced by the sensor itself.

[0038] In some embodiments, to avoid frequency band overlap, a filter bank is introduced, wherein the low-frequency disturbance estimate d` l Through a low-pass filter Gl ( s )= ωl / ( s + ωl Extraction: The bandwidth of the filter can be set according to the characteristics of the low-frequency observer, where... ωl The bandwidth of the low-frequency observer, for example ωl =5rad / s, s The Laplace operator; wherein the intermediate frequency disturbance estimate d` m Through bandpass filter Gm ( s )= The center frequency and damping ratio of the extracted data can be adjusted according to the characteristics of the intermediate frequency disturbance. ξm For the damping ratio, take ξm=0.7, ωm For the center frequency, take ωm =3 rad / s (corresponding to a center frequency of approximately 0.5 Hz); the estimated high-frequency disturbance value d` h Extracted using a high-pass filter (cutoff frequency 5Hz).

[0039] In some embodiments, to improve adaptability, an adaptive weight allocation is introduced, based on past performance. T The energy output of each observer within a second, for example, calculating the energy output of each observer over the past 2 seconds every 2 seconds. Pi = dτ The composite weights are dynamically adjusted, and the estimated total composite perturbation is d`= αl d` l + αm d` m + αh d` h The weighting coefficient αi = , i = l , m , h ,∥d` i ∥ represents the Euclidean norm of the estimated value. Under typical operating conditions, αl ≈0.4, αm ≈0.35, αh ≈0.25.

[0040] Experimental verification The test was conducted under the following conditions: rated load of 40 tons, rope length of 25m, and target speed of 50m / min. Steady-state positioning error ≤ ±2mm; The residual swing angle is ≤0.5°; For sinusoidal wind load disturbances with an amplitude of 200N and a frequency of 0.5Hz, the disturbance suppression ratio reaches 28dB; For a 20Hz, 100N impact disturbance caused by gear meshing, the recovery time is ≤0.25 s. Compared with the traditional single ESO method, the steady-state estimation error of this invention in the full-band disturbance environment is reduced from ±8% to ±2%, and the high-frequency impact recovery time is shortened by 58%.

[0041] In some embodiments, the crane closed-loop control method further includes constructing a digital twin high-fidelity dynamic model, such as establishing a high-fidelity dynamic model on an edge computing node (2TOPS computing power): M(q)q``+C(q,q`)q`+G(q)+F fric =u+dext In the formula, M(q) is the mass matrix dependent on the generalized coordinate vector q, q= ,in For the displacement of the trolley, For the load swing angle, The length of the rope. Let q'' be the elastic deformation of the wire rope, in meters (m), q'' be the generalized acceleration vector, C(q,q') be the Coriolis force and centrifugal force matrix, q' be the generalized velocity vector, G(q) be the gravity term vector, and F be the force vector. fric Let u be the friction force vector, and d be the control force vector. ext The external disturbance vector; The virtual model operates synchronously with the physical entity at a frequency of 100 Hz, receiving real-time state feedback (xt, θ, l) from the physical system. xt , θ , l ), and calculate the output y virtual .

[0042] Define the twin bias vector: e twin =y physical y virtual In the formula, y physical y is the actual output vector of the physical system. virtual The output vector of the digital twin virtual model; Establish a deviation prediction model: e` twin =A twin e twin +B twin Δp, where e` twin For twin bias vector e twin The first derivative with respect to time, A, is calculated using the difference approximation. twin Let B be the deviation state matrix. twin The input matrix is ​​the deviation matrix, and Δp is the unmodeled parameter deviation vector, including the change in friction coefficient and the degradation of wire rope stiffness; this embodiment mainly considers the load mass. ml Changes and steel wire rope rigidity kl Degradation. Through online identification of p= Update Δp.

[0043] Online identification of time-varying parameter vectors using the recursive least squares method: p( t )= ,in For load quality, For friction, Here is the stiffness coefficient of the wire rope; the parameter update rule is: P` ( k +1)=p`(k )+K RLS ( k )[ y ( k ) ( k )p`( k In the formula, p` ( k +1) is the first k +1 sampling time point parameter estimation vector, p`( k ) is the first k The parameter estimation vector at each sampling time point, K RLS ( k ) is the first k The recursive least-squares gain vector at each sampling time. y ( k ) is the first k The system output scalar at each sampling time. ( k ) is the first k Regression vector at each sampling time ( k The transpose of ) The update rule for the gain matrix is: K RLS ( k )= In the formula, p ( k -1) is the first k The covariance matrix at -1 sampling time point, λ This is the forgetting factor, with a value range of 0-1; The update rule for the covariance matrix is: P( k )= [P( k 1) K RLS ( k ) ( k )P( k 1)], where P( k ) is the first k The covariance matrix at each sampling time point is used to perform fusion correction on the intermediate frequency observer and the high frequency observer using twin bias: =d m +K twin e twin , =d h +K high e` twin In the formula, The corrected intermediate frequency disturbance estimate, d m K is the original intermediate frequency disturbance estimate. twin The twin bias correction gain matrix has a value of K. twin = I²×2, The corrected high-frequency disturbance estimate is given by the value of [value]. = I²×2, d h K is the original high-frequency disturbance estimate. high This is the high-frequency correction gain matrix; The corrected mid-frequency disturbance estimate and high-frequency disturbance estimates Replace the original estimate with the low-frequency disturbance estimate d` l The values ​​are synthesized into a total disturbance estimate. Experiments show that after adding twin correction, under the condition of sudden changes in load quality, the mid-frequency disturbance estimation error is further reduced by 40%, and the convergence time of high-frequency impulses is shortened from 0.1s to 0.06s.

[0044] Digital twin systems endow this control method with "online model adaptation" capabilities. When the actual load mass, friction, and other parameters of the crane vary within a wide range, this system does not "tolerate" this change, but rather "adapts" to it. It can always maintain an internal model that closely matches the actual situation, thereby potentially maintaining efficient control performance and increasing parameter perturbation tolerance by more than 66%. Furthermore, these system physical parameters (such as wire rope stiffness kl, friction coefficient Ff, etc.) are themselves important indicators of the crane system's health status. For example, monitoring the long-term trend of kl values ​​can determine whether the wire rope is fatigued or damaged; monitoring abnormal changes in Ff can determine whether the lubrication system has failed. This provides new model-based data support for achieving predictive maintenance and intelligent health management of cranes, elevating safety assurance from a passive response to a higher level of proactive prediction.

[0045] Furthermore, the primary task of digital twins is to track slowly time-varying physical parameters of a system (such as load mass and friction coefficient) through online parameter identification (e.g., recursive least squares). It addresses the fundamental problem of "model inaccuracy." Therefore, its correction information is mainly used to correct observers that are sensitive to model parameters. In this scheme, twin bias (e`) can be selected. twin The intermediate frequency (neural network) and high frequency (sliding mode) observers are corrected by using the low frequency ESO, which is designed to handle the "total disturbance" including parameter changes. It is more robust, so it is not directly corrected, thus avoiding redundancy and conflict of correction information.

[0046] Twin bias e`twin Instead of being directly superimposed on the observations, it is achieved through a carefully designed gain matrix (e.g., K). twin K high After adjustment, this gain is used as a correction term. The magnitude of this gain is crucial for ensuring system stability and avoiding oscillations. The range of the gain value can be determined using system stability theory (such as Lipnov stability analysis) to ensure that the correction process is convergent, rather than divergent or oscillating.

[0047] Online parameter identification algorithms (such as RLS) incorporate a "forgetting factor λ," ensuring that the parameter estimates are updated smoothly and gradually, rather than abruptly. This guarantees that the digital twin model itself evolves smoothly, and its output y... virtual There will be no mutation, thus avoiding twinning bias e` twin It is also a continuous and smooth signal. This smooth correction signal is fed into the observer, preventing sudden shocks to the system and thus avoiding oscillations.

[0048] In summary, through three major mechanisms—clear division of responsibilities, controllable gain of the correction channel, and smooth parameter updates—the collaborative work between the digital twin and multiple observers is ensured to be efficient and stable.

[0049] In some embodiments, the forgetting factor of the recursive least squares method λ The value is 0.98, and the initial covariance matrix P(0) = I3×3, where I3×3 is a 3x3 identity matrix, and the sampling interval is... Ts =0.01 seconds. These parameters were selected and optimized to balance the identification response speed and robustness to noise.

[0050] Regression vector ( k The system is constructed offline based on system dynamics.

[0051] Example of identification results: When the load mass suddenly increases from 20 tons to 30 tons, ml Within 0.5 seconds, it tracks to within 95% of the true value; when the wire rope stiffness decreases by 10% due to wear, kl It converges to the new value within 2 seconds.

[0052] In some embodiments, in the digital twin high-fidelity dynamic model, the elastic deformation of the wire rope... δ The dynamics are based on the stiffness coefficient kl and damping coefficient cl Confirmed, satisfied mlδ ``+ clδ `+ klδ = T ,in mlFor load quality, δ `` represents elastic acceleration cl The damping coefficient is... δ `is the elastic deformation velocity` kl This is the stiffness coefficient. T Let represent the tension in the wire rope. The elastic model of the wire rope is: mlδ`` + clδ `+ klδ = T In the formula, = 20000 kg, damping coefficient = 500 N s / m, stiffness coefficient kl =2× N / m, T The wire rope tension (N) is represented by this detailed modeling, which helps to more accurately reflect the actual dynamics of the system.

[0053] The overall control process is roughly as follows: 1. Initialization: Read the initial state of the crane and set the observer, filter, and RLS parameters; 2. Data Acquisition: Read data from the encoder, tilt sensor, and draw wire sensor at a frequency of 100 Hz, and update y; 3. Digital twin synchronization: Input the physical state q into the virtual model and calculate y. virtual and twin deviation e twin ; 4. Parameter identification: Run the RLS algorithm to update p, and then update Δp; 5. Three-band disturbance estimation: Low-frequency ESO update d` l ; Intermediate Frequency RBF Network Update d` m and with K twin e twin Corrected ; High-frequency sliding mode observer update d` h and with K high e˙ twin Corrected ; 6. Frequency Allocation and Synthesis: After filtering, the total disturbance d' is weighted and synthesized. 7. Sliding mode control law calculation: Calculate the control force based on the target trajectory and total disturbance estimate. u Output to the driver; 8. Repeat steps 2-7.

[0054] The control method provided by this invention can be widely applied to various types of bridge, gantry, and tower cranes, especially in scenarios such as automated docks and smart factories where high positioning accuracy, anti-sway performance, and anti-interference capabilities are required. The multi-band observer and digital twin module in this method can be packaged into a standardized software library, facilitating portability to different crane controller models.

[0055] Under the same crane parameters and operating conditions, the closed-loop control method for cranes provided in this application addresses: 1. Positioning recovery time after a sudden load change (20t→30t): 1.2s without digital twin and 0.6s with digital twin; 2. Steady-state swing angle when wire rope stiffness degrades by 10%: 0.8° without digital twin and 0.3° with digital twin; 3. Maximum swing angle under extreme wind load (15m / s): 2.5° without digital twin and 1.8° with digital twin; 4. Parameter identification convergence time: only applicable to cranes with digital twin, ranging from 0.5 to 2s.

[0056] It is worth noting that, through long-term work, we have discovered that in the industrial environment where crane systems operate under high loads for extended periods, some microscopic and imperceptible physical changes gradually accumulate. For example, the fiber optic cable connecting the physical system controller motherboard and the digital twin server may have internal defects at one of its connectors that are difficult to detect with the naked eye during manufacturing or installation. These defects may manifest as tiny scratches or bubbles on the fiber end face, or stress concentration points inside the connector. As cranes operate for extended periods in port environments, periodic changes in ambient temperature (such as diurnal temperature variations and seasonal changes) cause minute thermal expansion and contraction of the fiber optic material. Simultaneously, the slight vibrations of the crane itself, especially the mechanical vibrations generated when the trolley moves at high speeds on the track, also apply continuous mechanical stress to the fiber optic connector. These seemingly insignificant factors can cause the initial defects at the fiber optic connector to gradually expand. This expansion is not a sudden break, but a slow, gradual degradation process, leading to a decrease in the connection stability between the fiber cores. The direct consequence is that the optical signal may occasionally experience weak scattering or attenuation during transmission. This scattering and attenuation is intermittent and has a small amplitude, insufficient to trigger a communication interruption alarm, but enough to affect the signal quality.

[0057] As operating time increases, internal defects or damage at the fiber optic connectors continue to worsen. This ongoing degradation further degrades the stability of optical signal transmission, leading to a decrease in the physical system's output data stream (y). physicalOccasionally, minor, non-periodic packet loss or transmission delays may occur during transmission. For example, a sensor data packet that should have been transmitted completely within 0.01 seconds might occasionally lose a few bytes, or the entire packet might arrive a few microseconds later than expected. While these data transmission problems are not enough to cause a complete communication interruption, they do affect the speed at which the digital twin server receives data. physical The data stream exhibits subtle time misalignments or compromises in data integrity. For real-time control systems, such subtle time misalignments or data incompleteness are akin to the received information being "slightly out of order" or "missing a few key words." While the overall information remains recognizable, its accuracy is compromised.

[0058] The physical system output data stream, characterized by subtle timing discrepancies and data incompleteness, directly impacts the performance of the recursive least squares algorithm in the digital twin system. The recursive least squares algorithm relies on continuous and accurate input-output data streams to identify system characteristics online. When the data stream exhibits the aforementioned problems, the estimated values ​​of various system characteristics identified online by the recursive least squares algorithm, such as load mass, friction coefficient, and wire rope stiffness, begin to drift slightly and slowly when processing this "imperfect" data. This drift is not a drastic jump, but rather resembles a slowly deviating clock, gradually developing a small deviation between the indicated time and the actual time. For example, with the actual load mass remaining constant, the load mass estimate identified by the recursive least squares algorithm may slowly increase or decrease by a few percentage points over several hours. These drifting estimates can no longer accurately track the actual time-varying characteristics of the physical system, thus creating potential problems for subsequent control processes.

[0059] Due to drift in the estimated system characteristics identified online, the mismatch between the internal parameters of the digital twin dynamic model equations and the actual parameters of the physical system gradually increases when updating its own dynamic equations. For example, if the recursive least squares algorithm incorrectly estimates the load mass slightly higher than the actual value, the digital twin model will calculate the dynamic components such as the Coriolis force, centrifugal force matrix, and gravity term based on this incorrect load mass. This leads to a decrease in the output of the digital twin virtual model (y... virtual ) and the actual output of the physical system (y physical The deviation e calculated by the twin deviation calculation formula between them. twin It no longer purely reflects unmodeled dynamics and random perturbations. Instead, this e twin The signal contains more model mismatch error. This means that what should have been a clear reflection of the "unknown behavior of the system" is now contaminated by noise indicating that the "known model of the system is inaccurate".

[0060] This twin bias e includes additional model mismatch error. twin It was used for online correction of the estimation results of the mid-frequency perturbation observer (radial basis function neural network) and the high-frequency perturbation observer (sliding mode observer). Although the correction gain matrix (K twin K high The value of ) is carefully optimized to ensure system stability, but due to the correction signal e twin It no longer accurately reflects pure perturbation information; instead, it introduces erroneous corrections. For example, if e twin The signal contains "spurious" biases due to the digital twin model's overestimation of load quality. Upon receiving this correction signal, the intermediate frequency (IF) and high frequency (HF) observers will attempt to "compensate" for this non-existent disturbance. This leads to a decrease in the estimation accuracy of the IF and HF disturbance observers, making their estimations of the actual disturbances less accurate. Under certain operating conditions, this erroneous correction can even trigger slight, imperceptible control oscillations. These oscillations may be so small that they are difficult to identify directly in routine monitoring, yet they continuously deplete the system's control margin.

[0061] Ultimately, due to the inability of the mid-frequency and high-frequency disturbance observers to provide accurate disturbance estimation and correction, the crane control system's disturbance suppression capability across the entire 0-50Hz frequency band decreased. This directly resulted in the crane failing to meet design requirements for positioning accuracy and transient response speed during precise positioning or rapid response operations. For example, when precisely placing a container in a designated position, the crane trolley might exhibit unexpected slight swaying, leading to positioning errors exceeding ±2mm. When encountering sudden wind loads or performing rapid start-stop operations, the system exhibited unexpected swaying and vibration, increasing operational risks and requiring the crane to expend more energy to overcome these unnecessary vibrations. The entire control system degenerated from its original "active cognition" to a "biased passive estimation," significantly weakening its adaptive capability to time-varying parameters, thus affecting the crane's overall operating efficiency and safety in complex industrial environments.

[0062] Therefore, the focus shifts from passively responding to data transmission problems to proactively predicting and avoiding them. In some embodiments, a microscopic signal analysis unit is integrated into the digital twin server to continuously monitor the physical signal characteristics of the fiber optic transmission link, such as slight attenuation of the optical signal, signal-to-noise ratio fluctuations, and spectral changes. Based on these physical signal characteristics, a transmission channel status score is established, and a trend prediction algorithm is used to predict the future health status of the fiber optic medium. The predicted future status score is continuously compared with a preset warning line, triggering a tiered response. In cases of mild warning, the disturbance observer correction gain is automatically adjusted to reduce reliance on digital twin correction information. In cases of severe warning, a preventative maintenance alarm is issued, recommending manual intervention. This shift from monitoring "data content" to "physical status of the data channel" enables early warning and proactive intervention for potential data transmission problems, fundamentally ensuring the robustness of parameter identification in the digital twin system and the stability of high-precision crane control.

[0063] The specific method logic is as follows: 1. Optical Signal Feature Acquisition and Preliminary Processing: A dedicated signal analysis unit is set up at the receiving end of the digital twin server. This unit continuously collects various subtle features of the optical signal transmitted from the physical crane controller motherboard. These features include, but are not limited to, whether the intensity of the optical signal is slightly weakened, the amount of noise present in the signal, and whether there are slight reflections or dispersion phenomena during the transmission of the optical signal. The collected raw signal undergoes a "cleaning" process to remove some irrelevant interference, making the signal clearer for subsequent accurate analysis.

[0064] 2. Microscopic Feature Pattern Recognition: After initial cleaning, the signals are carefully analyzed to search for specific patterns hidden within them that are related to minute damage existing inside the fiber optic connection points. These patterns may manifest as increased signal fluctuations at certain frequencies or subtle changes in signal attenuation. The system pays particular attention to signal characteristics caused by periodic changes in port ambient temperature and crane vibrations, which lead to the gradual expansion of fiber optic defects. By identifying these unique signal patterns, the system can "detect" the microscopic physical changes occurring at the fiber optic connection points.

[0065] 3. Transmission Channel Status Assessment: Based on the identified microscopic feature patterns, the system calculates a "transmission channel status score." This score is like a health check report for the fiber optic connection point; a higher score indicates better physical condition of the fiber, while a lower score suggests potential problems or degradation trends. This score reflects the health status of the fiber optic transmission channel in real time.

[0066] 4. Future State Prediction: The controller system utilizes current and past "transmission channel status scores" and employs a prediction method based on the gradual changes in optical fiber materials over long-term operation to forecast how the state of the optical fiber connection point will change in the next few hours or days. This prediction can anticipate the potential extent of optical fiber degradation, providing a time window for subsequent countermeasures.

[0067] 5. Early Warning and Response Mechanism: The controller system will continuously compare the calculated future "transmission channel status score" with several preset warning thresholds, and trigger different response measures based on the comparison results: a. First warning line (mild warning): If the calculation results show that the fiber optic status is about to reach the mild warning line, indicating that the data transmission quality may begin to be slightly affected in the near future, the controller system will immediately notify the relevant modules. For example, it will suggest that the module responsible for adjusting the disturbance compensation level temporarily adopt a more conservative strategy, slightly reduce its reliance on the correction information provided by the digital twin system, and prepare for the possible decline in data quality; b. Second warning line (severe warning): If the calculation results show that the fiber optic condition is about to reach the severe warning line, indicating that the data transmission quality may soon have a significant impact on the precise control of the crane, the system will immediately issue a "preventive maintenance" alarm to the operator. This alarm will explicitly recommend checking the fiber optic connection points, and even considering replacing them in advance, thus intervening before the data transmission problem actually affects the precise control performance of the crane.

[0068] Traditional system health monitoring typically focuses on equipment operating status or data output results. The unique aspect of this solution lies in its shift of focus from "data" to the microscopic physical state of the "data channel," utilizing these extremely subtle physical signal changes to predict future data problems. It doesn't address issues after they occur, but rather anticipates and intervenes at the nascent stage. This shift from "outcome-oriented" to "root cause prediction" demonstrates a profound insight into the sources of system vulnerability and dares to seek clues to solving control system problems at seemingly unrelated physical levels. The value of this solution lies in its non-linear approach. It doesn't directly solve the recursive least squares parameter drift problem, but rather triggers preventative maintenance or system adaptive calibration before problems occur by predicting fiber optic degradation. This means that the recursive least squares algorithm is not exposed to heavily polluted data streams for extended periods, fundamentally ensuring its robustness and the accuracy of the digital twin model. This radical solution not only addresses explicit problems but also resolves implicit, long-term accumulated system health risks, improving the overall operating efficiency and safety of the crane system. It elevates security from a passive response to a higher level of proactive prediction, which is a manifestation of innovative thinking.

[0069] Therefore, by adopting this solution, the crane's closed-loop control method / system can proactively detect and predict potential degradation of the data transmission fiber, thus taking preventative measures before data transmission issues substantially impact control performance. This effectively avoids parameter estimation drift caused by data contamination in the recursive least squares algorithm of the digital twin system, ensuring the accuracy of the digital twin dynamics model. Consequently, it maintains the crane's disturbance suppression capability across the 0-50Hz full-frequency band and its ±2mm high-precision positioning performance, significantly improving the crane's operational stability and reliability in heavy-load, high-precision positioning scenarios.

[0070] A second aspect of this application provides a closed-loop control system for a crane, used to implement the closed-loop control method for a crane as described above. The control system includes: The multi-band adaptive disturbance observer building module is used to construct a multi-band adaptive disturbance observer, which decomposes the composite disturbance vector d into low-frequency component vectors di. l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h ; The low-frequency component vector estimation submodule is used to estimate the low-frequency component vector using an extended state observer; The intermediate frequency component vector estimation submodule is used to estimate the intermediate frequency component vector using an adaptive radial basis function neural network; The high-frequency component vector estimation submodule is used to estimate the high-frequency component vector using a high-order sliding mode observer based on the superspiral algorithm; The fusion output module is used to convert the low-frequency disturbance estimate d` l Mid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

[0071] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the crane closed-loop control method as described above.

[0072] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the invention. The advantages of the present invention have been fully and effectively realized. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments; any variations or modifications can be made to the implementation of the present invention without departing from these principles.

Claims

1. A closed-loop control method for cranes, characterized in that, Includes the following steps: Construct a multi-band adaptive disturbance observer to decompose the composite disturbance vector d into low-frequency component vector d l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h ,in: The low-frequency component vector is estimated using an extended state observer, and the state estimation equation is as follows: =A e z+B eu +L l (y C e z), where, To extend the first derivative of the state estimate z with respect to time, A e To expand the system matrix, z=[x` T , d` l T ] T Where x' is the system state estimate, d' l B is the estimated value for low-frequency disturbance. e To expand the input matrix, u L is the driving force of the motor. l Let y be the gain matrix of the low-frequency observer, and y be the output vector of the actual system. e To expand the output matrix, the low-frequency disturbance estimate d` l The update pattern is ` l = ωl d` l + ωl L l (y y`), where, ` l The low-frequency disturbance estimate d` l The first derivative with respect to time, ωl Here, represents the bandwidth of the low-frequency observer, in rad / s; y is the actual system output vector; y' is the estimated system output value; and y' = C. e z; An adaptive radial basis function neural network is used to estimate the intermediate frequency component vector. The output of the neural network is: d` m =W T Φ(x), where d` m W is the estimated value of the intermediate frequency disturbance. T Let W be the transpose of the network weight matrix W, where W has dimension 1. N ×2, where N Let denoted as the number of neurons, and Φ(x) be the Gaussian kernel vector with dimension . N ×1, its first i The components are: i (x) ,in i (x) is the first i The output of each neuron It is a natural exponential function. Let the system state vector x and the first i vector of Gaussian kernel center point The difference in Euclidean distance when laid flat, For the first i The width of a Gaussian kernel, where the adaptive update rule of the weight matrix W is: =ΓΦ(x) K m W, where, Let W be the first derivative of the weight matrix W with respect to time, and Γ be the adaptive gain matrix. N × N A positive definite diagonal matrix, Let e ​​be the observation error vector m =y The transpose of y', K m for The coefficient matrix of the correction term is N × N matrix; A high-order sliding mode observer based on the superspiral algorithm is used to estimate the high-frequency component vector, which has the following continuous form: h = λ 1 sgn(e)+ λ 2∫sgn(e) dt In the formula, h The high-frequency disturbance estimate d` h The first derivative with respect to time, λ 1 represents the sliding mode gain, used to control the convergence speed. Let the observation error vector be e=y Let y' be a vector formed by the square roots of the absolute values ​​of each component, and sgn(e) be the coincidence function of the observation error vector, where each component takes the value +1, -1, or 0. λ 2 represents the integral gain, ∫sgn(e) dt The discretized form of the high-frequency observer is the integral of the sign function vector over time: d` h ( k +1)=d` h ( k )+ Ts [ λ 1 sgn(e( k ))+ λ 2v( k )],v( k +1)=v( k )+ Ts sgn(e( k In the formula, d` h ( k +1) indicates the first k +1 sampling time point high-frequency disturbance estimate, d` h ( k ) is the first k The high-frequency disturbance estimate at each sampling time. Ts The sampling period is λ 1 represents the sliding mode gain. For the first The square root vector of the absolute value of the observation error at each time point, sgn(e( k )) is the first k The sign function vector of the observation error at each sampling time. λ 2 represents the integral gain, v( k ) is the first k The integral state vector at each sampling time point, v( k +1) is the first k The integral state vector at +1 sampling time points; The low-frequency disturbance estimate d` l Mid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

2. The closed-loop control method for cranes as described in claim 1, characterized in that, The low-frequency component has a frequency range of 0-0.5Hz, the mid-frequency component has a frequency range of 0.5-5Hz, and the high-frequency component has a frequency range of 5-50Hz.

3. The crane closed-loop control method as described in claim 2, characterized in that, The low-frequency disturbance estimate d` l Through a low-pass filter Gl ( s )= ωl / ( s + ωl Extract, where ωl The bandwidth of the low-frequency observer is... s The Laplace operator; the estimated mid-frequency disturbance d` m Through bandpass filter Gm ( s )= Extract, of which ξm For the damping ratio, ωm The center frequency; the estimated high-frequency disturbance value d` h Extracted using a high-pass filter.

4. The closed-loop control method for cranes as described in claim 1, characterized in that, Based on the past T Energy output by each observer within seconds Pi = dτ The composite weights are dynamically adjusted, and the estimated total composite perturbation is d`= αl d` l + α m d` m + αh d` h The weighting coefficient αi = , i = l , m , h ,∥d` i ∥ represents the Euclidean norm of the estimated value.

5. The crane closed-loop control method according to any one of claims 1 to 4, characterized in that, It also includes constructing a high-fidelity dynamic model for digital twins: M(q)q``+C(q,q`)q`+G(q)+F fric =u+d ext In the formula, M(q) is the mass matrix dependent on the generalized coordinate vector q, q= ,in For the displacement of the trolley, For the load swing angle, The length of the rope. Let q'' be the elastic deformation of the wire rope, q'' be the generalized acceleration vector, C(q,q') be the Coriolis force and centrifugal force matrix, q' be the generalized velocity vector, G(q) be the gravity term vector, and F be the force vector. fric Let u be the friction force vector, and d be the control force vector. ext The external disturbance vector; Define the twin bias vector: e twin =y physical y virtual In the formula, y physical y is the actual output vector of the physical system. virtual The output vector of the digital twin virtual model; Establish a deviation prediction model: e` twin =A twin e twin +B twin Δp, where e` twin For twin bias vector e twin The first derivative with respect to time, A twin Let B be the deviation state matrix. twin The deviation input matrix is ​​Δp, which is the unmodeled parameter deviation vector, including the change in friction coefficient and the degradation of wire rope stiffness. Online identification of time-varying parameter vectors using the recursive least squares method: p( t )= ,in For load quality, For friction, Here is the stiffness coefficient of the wire rope; the parameter update rule is: P` ( k +1)=p`( k )+K RLS ( k )[ y ( k ) ( k )p`( k In the formula, p` ( k +1) is the first k +1 sampling time point parameter estimation vector, p`( k ) is the first k The parameter estimation vector at each sampling time point, K RLS ( k ) is the first k The recursive least-squares gain vector at each sampling time. y ( k ) is the first k The system output scalar at each sampling time. ( k ) is the first k Regression vector at each sampling time ( k The transpose of ) The update rule for the gain matrix is: K RLS ( k )= In the formula, p ( k -1) is the first k The covariance matrix at -1 sampling time point, λ This is the forgetting factor, with a value range of 0-1; The update rule for the covariance matrix is: P( k )= [P( k 1) K RLS ( k ) ( k )P( k 1)], where P( k ) is the first k The covariance matrix at each sampling time point is used to perform fusion correction on the intermediate frequency observer and the high frequency observer using twin bias: =d m +K twin e twin , =d h +K high e` twin In the formula, The corrected intermediate frequency disturbance estimate, d m K is the original intermediate frequency disturbance estimate. twin This is the twin bias correction gain matrix. The corrected high-frequency disturbance estimate, d h K is the original high-frequency disturbance estimate. high This is the high-frequency correction gain matrix; The corrected mid-frequency disturbance estimate and high-frequency disturbance estimates Replace the original estimate with the low-frequency disturbance estimate d` l The values ​​are combined to form a total disturbance estimate.

6. The crane closed-loop control method as described in claim 5, characterized in that, The forgetting factor of the recursive least squares method λ The value is 0.98, and the initial covariance matrix P(0) = I3×3, where I3×3 is a 3x3 identity matrix, and the sampling interval is... Ts =0.01 seconds.

7. The crane closed-loop control method as described in claim 5, characterized in that, In the digital twin high-fidelity dynamic model, the elastic deformation of the wire rope δ The dynamics are based on the stiffness coefficient kl and damping coefficient cl Confirmed, satisfied mlδ ``+ clδ `+ klδ = T ,in ml For load quality, δ `` represents elastic acceleration cl The damping coefficient is... δ `is the elastic deformation velocity` kl This is the stiffness coefficient. T This refers to the tension of the steel wire rope.

8. The closed-loop control method for a crane as described in claim 5, characterized in that, A micro-signal analysis unit is integrated on the digital twin server to continuously monitor the physical signal characteristics of the optical fiber transmission link. Based on the physical signal characteristics, a transmission channel status score is established, and a trend prediction algorithm is used to predict the future health status of the optical fiber medium. The predicted future status score is continuously compared with a preset warning line, and a graded response is triggered. In the case of a mild warning, the disturbance observer correction gain is automatically adjusted to reduce the dependence on digital twin correction information. In the case of a severe warning, a preventive maintenance alarm is issued.

9. A crane closed-loop control system, used to implement the crane closed-loop control method as described in claim 1, characterized in that, include: The multi-band adaptive disturbance observer building module is used to construct a multi-band adaptive disturbance observer, which decomposes the composite disturbance vector d into low-frequency component vectors di. l , intermediate frequency component vector d m and high-frequency component vector d h , satisfying d=d l +d m +d h ; The low-frequency component vector estimation submodule is used to estimate the low-frequency component vector using an extended state observer; The intermediate frequency component vector estimation submodule is used to estimate the intermediate frequency component vector using an adaptive radial basis function neural network; The high-frequency component vector estimation submodule is used to estimate the high-frequency component vector using a high-order sliding mode observer based on the superspiral algorithm; The fusion output module is used to convert the low-frequency disturbance estimate d` l Mid-frequency disturbance estimate d` m and high-frequency disturbance estimate d` h Combined into a total disturbance estimate: d`=d` l +d` m +d` h The total disturbance estimate d` is then output as a feedforward compensation to the crane controller for control.

10. A computer-readable storage medium storing instructions, characterized in that, When the instructions are executed by the processor, the processor is configured to perform the crane closed-loop control method as described in any one of claims 1 to 8.

Images