Tower crane adaptive neural network control method, system, medium and equipment

Abstract

The present invention relates to the technical field of tower crane control. The present invention discloses a tower crane adaptive neural network control method, system, medium and equipment, including: based on tower crane system parameters, using an adaptive neural network as a controller, driving auxiliary vectors, drivable error vectors and under-actuated state vectors to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, combines the control input caused by the dead zone characteristic, obtains the control input of the tower crane system, and drives the state vector of the tower crane system to track the desired position; wherein the drivable error vector is the error between the desired position and the drivable state vector, the drivable state vector includes the cantilever rotation angle and the trolley displacement; the under-actuated state vector is the load swing angle; the auxiliary vector represents the coupling relationship between the drivable and under-actuated states. The robustness of the tower crane system is greatly enhanced, and the stability of the tower crane system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tower crane control, and in particular to a tower crane adaptive neural network control method, system, medium and equipment. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The main challenge facing the control of tower crane systems comes from their underactuated nature, that is, the number of control inputs is less than the number of degrees of freedom to be controlled. This can lead to uncontrolled load swings, which may affect safety and accuracy. To address this problem, researchers have developed a variety of different control strategies, including adaptive control, sliding mode control, and intelligent control methods such as fuzzy logic and neural networks. Adaptive control methods are particularly effective, as they are able to adjust control parameters in real time to cope with changing loads and environmental conditions; sliding mode control is another widely studied robust control technique that has attracted attention for its insensitivity to parameter changes and external disturbances; intelligent control methods, such as fuzzy logic and neural networks, have also been successfully applied to tower crane systems. These methods do not require an accurate mathematical model of the tower crane system, which is advantageous when designing complex and nonlinear dynamics, and they provide a means to more effectively deal with uncertainty and nonlinearity.

[0004] In view of the above discussion on tower crane system control, the unresolved issues and potential directions for future research are summarized as follows:

[0005] 1) Most existing tower crane control methods are developed using simplified models that linearize or approximate the dynamics of the crane. If the state variables of the tower crane system deviate significantly from the equilibrium point, perhaps due to unforeseen disturbances, there may be a large difference between these simplified models and the actual system dynamics, and this difference may adversely affect the control performance and may even cause system instability.

[0006] 2) In real-world servo motors, input dead band is common practice, which is an inherent characteristic of electromechanical systems including tower crane systems. However, in order to simplify the design of tower crane controllers, the effect of dead band is usually ignored, and this oversight may degrade the performance of the control system and may even cause instability.

[0007] 3) Existing full-drive system methods can only ensure the eventual uniform boundedness of the equilibrium point of the complex tower crane system, which means that the error vector will eventually converge to a very small area, although not exactly at the origin.

[0008] 4) Repeated differentiation in order to obtain unavailable higher-order signals may introduce noise and distort the signal, which may reduce the accuracy of the feedback loop and may cause excessive transient control inputs.

[0009] 5) For underactuated tower crane systems subject to uncertain dynamics, state unavailability, and input dead zones, there is an obvious need for a control approach that can ensure that the unactuated states are driven to the target values ​​and has a solid theoretical foundation. This approach should also be easily adaptable to various operational requirements and aim to simplify the derivation process. However, to date, no comprehensive control framework has been proposed that successfully integrates accuracy, adaptability, and wide applicability. Summary of the invention

[0010] In order to solve the above problems, the present invention provides an adaptive neural network control method, system, medium and equipment for a tower crane, which integrates an adaptive neural network into a controller to avoid problems such as parameter uncertainty and input dead zone, thereby greatly enhancing the robustness of the tower crane system. By carefully constructing auxiliary variables that reflect the coupling relationship between the driven and undriven states, the original nonlinear under-driven closed-loop system is reshaped into a linear full-driven system. This conversion ensures the collective asymptotic convergence of the auxiliary variables and the driven and undriven states, thereby improving the stability of the tower crane system.

[0011] In order to achieve the above object, the present invention adopts the following technical solution:

[0012] A first aspect of the present invention provides a tower crane adaptive neural network control method, which comprises:

[0013] Obtain tower crane system parameters and desired position;

[0014] Based on the tower crane system parameters, an adaptive neural network is used as a controller to drive the auxiliary vector, the drivable error vector and the under-actuated state vector to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, based on the control input generated by the controller and combined with the control input caused by the dead zone characteristic, the control input of the tower crane system is obtained, and the state vector of the tower crane system is driven to track the desired position;

[0015] Among them, the driveable error vector is the error between the desired position and the driveable state vector, and the driveable state vector includes the cantilever rotation angle and the trolley displacement; the under-actuated state vector is the load swing angle; and the auxiliary vector represents the coupling relationship between the driveable and under-actuated states.

[0016] Furthermore, the state vector of the tower crane system includes a cantilever rotation angle, a trolley displacement and a load swing angle.

[0017] Furthermore, the tower crane system parameters include the mass of the trolley and the load, the moment of inertia of the cantilever, the length of the suspension rope and the friction coefficient.

[0018] Furthermore, the auxiliary vector is constructed based on the dynamic properties of the underactuated subsystem in the tower crane system, and the auxiliary vector is expressed as: ,in, is a generalized vector, for The first derivative with respect to time is for The second derivative with respect to time is The expression is = , and denote the drivable state vector and the underactuated state vector respectively, and represents a positive diagonal matrix, is the drivable error vector, , , , , , , , , , , , , , , , , , , , , , is the load mass, is the length of the suspension rope, and is the friction coefficient, g is the acceleration due to gravity, is the cantilever rotation angle, is the trolley displacement, and is the load swing angle, , , and They are , , and Abbreviation of .

[0019] Furthermore, the auxiliary vector is estimated by a state observer, and the state observer is: ,in, is the auxiliary vector The estimate, as well as Represent the input weight vector and estimation error The estimated value of is the observation gain matrix, , , , , is a known nominal diagonal matrix, is the basis function, express The estimate, contains a state vector that cannot be measured directly, Output for the system.

[0020] Furthermore, the adaptive neural network is: ,in, represents the control matrix to be adjusted, is a known nominal diagonal matrix, is the basis function, express The estimate, contains a state vector that cannot be measured directly, is the auxiliary vector, as well as Represent the input weight vectors and estimation error An estimate of is obtained from the update rate.

[0021] Furthermore, the tower crane system adopts a dynamic model and is decomposed into two subsystems through a full drive system method: , ,in, and denote the drivable state vector and the underactuated state vector respectively, is the control input vector, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , and are the masses of the trolley and the load, respectively. is the cantilever moment of inertia, is the length of the suspension rope, To control the torque of the cantilever swing, is the translation control force of the trolley, , , and is the friction coefficient, g is the acceleration due to gravity, is the cantilever rotation angle, is the trolley displacement, and is the load swing angle, , , , , and They are , , , , and Abbreviation of .

[0022] A second aspect of the present invention provides a tower crane adaptive neural network control system, comprising:

[0023] A data acquisition module, configured to: acquire tower crane system parameters and a desired position;

[0024] A control module is configured to: based on the tower crane system parameters, use an adaptive neural network as a controller to drive the auxiliary vector, the drivable error vector and the under-actuated state vector to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, based on the control input generated by the controller, combined with the control input caused by the dead zone characteristic, obtain the control input of the tower crane system, and drive the state vector of the tower crane system to track the desired position;

[0025] Among them, the driveable error vector is the error between the desired position and the driveable state vector, and the driveable state vector includes the cantilever rotation angle and the trolley displacement; the under-actuated state vector is the load swing angle; and the auxiliary vector represents the coupling relationship between the driveable and under-actuated states.

[0026] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor and implements the steps in the adaptive neural network control method for a tower crane as described above when the program is executed by the processor.

[0027] The fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the program, the steps in the adaptive neural network control method for tower crane as described above are implemented.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention provides an adaptive neural network control method for a tower crane, which integrates an adaptive neural network into a controller to avoid problems such as parameter uncertainty and input dead zone, thereby greatly enhancing the robustness of the tower crane system. In addition, by carefully constructing auxiliary variables that reflect the coupling relationship between driven and undriven states, the original nonlinear under-driven closed-loop system is reshaped into a linear full-driven system. This conversion ensures the collective asymptotic convergence of the auxiliary variables and the driven and undriven states, thereby improving the stability of the tower crane system.

[0030] The invention provides an adaptive neural network control method for a tower crane, which uses a model-free state observer to accurately estimate unmeasurable states and compensate unknown dynamics online, thereby avoiding the problems of repeated numerical differentiation or discontinuous robust terms. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which constitute a part of the specification of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute limitations of the present invention.

[0032] Figure 1It is a schematic diagram of a freedom tower crane system according to the first embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of a tower crane test platform according to Embodiment 1 of the present invention;

[0034] Figure 3 This is a comparative test diagram of Experiment 1 of Example 1 of the present invention;

[0035] Figure 4 This is a diagram of experimental results of the control method proposed in Experiment 2 of Example 1 of the present invention for different load masses;

[0036] Figure 5 This is a graph showing experimental results of the control method according to the first embodiment of the present invention for different rope lengths;

[0037] Figure 6 This is a diagram of experimental results of the control method proposed in Example 1 of the present invention for different disturbances. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] In the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention is further described below with reference to the accompanying drawings and embodiments.

[0041] Terminology explanation:

[0042] The EEMA control method, whose full name is end-effector motion-based approach (EEMA), comes from the paper New energy analytical results for the regulation of underactuated overhead cranes: an end-effector motion-based approach, which was published in the journal IEEE Transactions on Industrial Electronics, Vol. 59, No. 5, in December 2012.

[0043] Embodiment 1

[0044] The purpose of this first embodiment is to provide an adaptive neural network control method for a tower crane.

[0045] Inspired by the full drive system approach, this embodiment designs a novel adaptive neural network control method for a 4-DOF tower crane system with parameter uncertainty, unmeasurable state variables and input dead zone. First, auxiliary variables containing the drive / under-actuated states and their derivatives and proportional differential terms are constructed to reconfigure the nonlinear under-actuated system into a linear full drive system, avoiding the requirement for any linearization operation. In the proposed control method, a radial basis function neural network is used to deal with the problems of parameter uncertainty and input dead zone. The state variables that cannot be directly measured are estimated by the established state observer, which avoids the problem of discontinuity and improves the accuracy of compensation and positioning. The asymptotic convergence of the carefully constructed auxiliary variables in this embodiment ensures that both the drive and under-actuated states can accurately reach their expected values. This embodiment provides the first continuous neural network adaptive control solution based on the full drive system approach for the under-actuated tower crane system, and a series of experimental results confirm the effectiveness and advantages of the proposed control method.

[0046] The tower crane adaptive neural network control method provided in this embodiment utilizes the characteristics of the full drive system to design a new neural adaptive control strategy for the tower crane system. It is not constrained by the original model. The carefully constructed auxiliary variables describe the entire under-actuated tower crane system through a linear model to ensure that the driven and undriven errors approach zero in an asymptotic manner. The main contributions are summarized as follows:

[0047] 1) In the process of designing the controller and conducting theoretical analysis of the tower crane system, there is no need to linearize the equilibrium point of the inherently complex nonlinear dynamics. Therefore, the effectiveness of the proposed controller can be guaranteed even when large load swing angles are caused by adverse external disturbances.

[0048] 2) In actual operation, the tower crane system always inevitably encounters problems such as parameter uncertainty and input dead zone, which may significantly reduce the control performance. In order to meet these challenges, this embodiment integrates the radial basis function neural network into the controller to comprehensively solve these problems, thereby greatly enhancing the robustness of the system.

[0049] 3) The original nonlinear underactuated closed-loop system is reshaped into a linear fully-actuated system by carefully constructing auxiliary variables that reflect the coupling relationship between the driven and unactuated states. This transformation ensures the collective asymptotic convergence of the auxiliary variables and the driven and unactuated states.

[0050] 4) We propose to use a model-free state observer to accurately estimate the unmeasurable states and compensate for the unknown dynamics online, avoiding the problems of repeated numerical differentiation or discontinuous robust terms.

[0051] The tower crane adaptive neural network control method provided in this embodiment includes the following steps:

[0052] Step 1: Construct a DOF tower crane system model.

[0053] Consider a 4-DOF tower crane system such as Figure 1 As shown, its dynamic model can be described as follows:

[0054] (1).

[0055] in, is the state vector, express The first derivative of express The second derivative of is a positive definite symmetric inertia matrix, is the centripetal-Coriolis force matrix, is the gravity vector, is the diagonal matrix containing the friction-related system, represents the original control input vector calculated by the control method, is the final control input vector.

[0056] in, and The specific formula is as follows:

[0057] ;

[0058] ;

[0059] ;

[0060] ;

[0061] ;

[0062] .

[0063] in, as well as The specific expression is: , , , , , , , , , , , , , , , , , , , , , , , , , .

[0064] For brevity and readability, Table 1 gives the definitions of system parameters.

[0065] Table 1. System parameters

[0066]

[0067] To adopt the full drive system approach, the tower crane system (i.e., formula (1)) is decomposed into the following two subsystems:

[0068] (2);

[0069] (3).

[0070] in, as well as They represent the drivable state vector and the under-actuated state vector respectively.

[0071] in, , , , , , , , , , , .

[0072] In practical applications, input dead zone is a common phenomenon in servo motors, which will directly affect the actual control input. It can be described as:

[0073] ; (4).

[0074] in, yes The lower bound of the dead zone, yes Dead zone lower bound, yes The upper bound of the dead zone, yes Dead zone upper bound, , , and Represents the unknown function related to the dead zone characteristics.

[0075] To simplify the controller design process, the control input vector of the tower crane system is divided into the following two parts:

[0076] (5).

[0077] in, Represents the control input caused by the deadband characteristic.

[0078] Based on the structure of formula (3), a generalized vector of the following form is constructed :

[0079] (6).

[0080] in, for The first derivative with respect to time is for The second derivative with respect to time is The expression is:

[0081] = (7).

[0082] in, represents a positive diagonal matrix, is the positioning error vector of the drivable state (drivable error vector), is the target vector of the drivable state. It should be noted that the auxiliary vector is constructed by incorporating the dynamic properties of the underactuated subsystem described in equation (3), thereby eliminating the need for additional control inputs.

[0083] It is not difficult to see is positive definite symmetric, therefore, from equation (7) we can know that:

[0084] (8).

[0085] Substituting formula (8) into formula (2), we can get:

[0086] = = (9).

[0087] in, represents an unknown function; ; , = Indicates the uncertainty factors of tower crane system; >0 is a known nominal diagonal matrix. Considering, or , so it can be seen that the tower crane system (i.e., formula (9)) is a full-drive system.

[0088] Solving the differential of equation (9) with respect to time, we can obtain:

[0089] (10).

[0090] Assumption 1: The load is always under the trolley / cantilever, that is:

[0091] (11).

[0092] Step 2: Determine the control objectives.

[0093] This embodiment studies the full-drive neural network control method of the uncertain tower crane system, and the control objectives can be divided into the following two parts:

[0094] Control Objective 1: The designed controller can still drive the trolley / cantilever to its target position in the presence of system uncertainty, unmeasurable state vectors, and input dead zone, that is:

[0095] (12).

[0096] in, express The target position vector (desired position), as well as Respectively as well as Target angle and position.

[0097] Control Objective 2: The designed controller can suppress and eliminate load swings, that is:

[0098] (13).

[0099] Step 3: Prior knowledge.

[0100] In order to facilitate the subsequent controller design, it is necessary to give some prior knowledge.

[0101] Lemma 1: Definition satisfy:

[0102] (14).

[0103] in, express A The i-th characteristic root of , n represents the dimension, Re represents a real number, represents a positive number, then there exists a positive definite symmetric matrix , so that:

[0104] (15).

[0105] Lemma 2: For , there exists a matrix satisfy:

[0106] (16).

[0107] Then, according to Lemma 1, when equation (16) satisfies, there exists a positive definite symmetric matrix , so that:

[0108] (17).

[0109] in,

[0110] (18).

[0111] in, Represents the control gain matrix to be adjusted.

[0112] Step 4: Design of adaptive neural network controller based on full drive system approach.

[0113] All-wheel drive system uncertainty It can be approximated by the following radial basis neural network structure:

[0114] ; ; (19).

[0115] in, represents a bounded input weight vector, is a basis function, and there is , and Respectively and The estimate, represents the estimation error, as well as satisfy: ,in, and Respectively and The upper bound of express The elements in express The elements in .

[0116] Inspired by the all-wheel drive system method, this embodiment designs a novel adaptive neural network control method to quickly drive the auxiliary vector and the drive error vector , underactuated state vector Converges to 0, the designed adaptive neural network method based on the full drive system method is:

[0117] (20).

[0118] in, represents the control matrix to be adjusted, as well as Respectively and The estimated value of can be obtained by the following update rate:

[0119] (twenty one);

[0120] (twenty two).

[0121] in, , , as well as represents the positive definite diagonal control matrix, represents a positive definite symmetric matrix The third line of elements, represents a positive definite symmetric matrix The sixth line of elements, represents a positive definite symmetric matrix, and Respectively and The upper bound of is: , .

[0122] Integrating equation (20) with respect to time, we can obtain:

[0123] (twenty three).

[0124] in, .

[0125] From equations (20), (23), (9) and (10), we can get:

[0126] (twenty four).

[0127] in, , , as well as Respectively , , , The estimation error.

[0128] From Lemma 2, we can see that the control gains in equations (20)-(23) must satisfy the following conditions:

[0129] ; (25).

[0130] in, are two positive real numbers, The expression is:

[0131] (26).

[0132] in, represents the control gain matrix to be determined, The definition of will be given later.

[0133] Theorem 1: The designed controller (23) and update rates (21) and (22) can ensure that the auxiliary vector Asymptotically converges to 0; at the same time, the cantilever rotation angle And trolley displacement It can converge to its target position asymptotically, and the load swing angle is suppressed and eliminated. Its mathematical description is:

[0134] ; ; (27).

[0135] Step 4: Design of state observer.

[0136] As shown in formula (19), contains several state vectors that cannot be directly measured. Therefore, this embodiment aims to design a state observer to accurately estimate .

[0137] From equations (9), (10) and (19), we can get:

[0138] (28).

[0139] in, , , , , , is the output of the system shown in formula (28), From equation (28), we can see that only and Can be measured directly, It cannot be obtained directly. Therefore, a state observer of the following form is designed to estimate :

[0140] (29).

[0141] in, for The estimate, and From equations (21) and (22), we can obtain is the observation gain matrix.

[0142] From equations (28) and (29), we can get the following estimation error system:

[0143] (30).

[0144] in, represents the estimation error vector.

[0145] Formula (30) can be written as follows:

[0146] (31).

[0147] definition:

[0148] (32).

[0149] Step 5: Stability analysis and convergence proof.

[0150] (1) Auxiliary vector and the estimated error vector The convergence of .

[0151] To prove that the designed method can ensure the auxiliary vector Asymptotically converges to 0, constructing a positive definite function of the following form for:

[0152] (33).

[0153] The derivative of formula (33) with respect to time is:

[0154] = + ≤ - - ≤ - (34).

[0155] Next, to prove Convergence of , construct a positive definite symmetric matrix of the following form for:

[0156] (35).

[0157] Taking the derivative of both ends of equation (35) with respect to time and substituting the results of equations (25) and (31) into it, we can obtain:

[0158] = ≤ - ≤ - (36).

[0159] Then, the constructed Lyapunov function is:

[0160] = (37).

[0161] Taking the derivative of both ends of equation (37) with respect to time, we can obtain:

[0162] = + - ≤ - - ≤ (38).

[0163] This shows that:

[0164] (39).

[0165] as well as and Asymptotically converges to 0, that is:

[0166] (40).

[0167] From equations (37) and (39), we can derive:

[0168] = ; = (41).

[0169] (2) Drive status and The convergence of .

[0170] Next, in order to prove the convergence of the drivable state, a positive definite Lyapunov candidate function of the following form is constructed:

[0171] (42).

[0172] Taking the derivative of both ends of equation (42) with respect to time, we can obtain:

[0173] = = = (43).

[0174] Among them, the derivation process used nature.

[0175] Substituting formula (40) into formula (43), we have:

[0176] (44).

[0177] From Barbara's lemma we know that:

[0178] (45).

[0179] From (7), (39) and (44), we can get:

[0180] (46).

[0181] From formula (7), we can easily get:

[0182] ; ; (47).

[0183] From equations (40), (44) and (45), we can get:

[0184] (48).

[0185] Then, by extending Barbara's lemma, we can get:

[0186] (49).

[0187] (3) Under-driven state and The convergence of .

[0188] To prove the underactuated state and The convergence of , constructs the following form of Lyapunov candidate function:

[0189] (50).

[0190] Taking the derivative of both ends of equation (50) with respect to time and substituting the result of equation (7) into it, we can obtain:

[0191] = + = = - = (51).

[0192] in, .

[0193] It is easy to see from formula (51) that when hour, ,So Decreasing, and:

[0194] (52).

[0195] According to equations (8), (39), (44), and (52), we can directly obtain:

[0196] (53).

[0197] Next, we transform Treated as a virtual input, and analyze the "0 input" dynamics independently (ie: ):

[0198] (54).

[0199] Integrating both ends of equation (54) with respect to time, we can obtain:

[0200] (55).

[0201] From formula (55), we can get:

[0202] (56).

[0203] According to the extended Barbara lemma, we can directly get:

[0204] (57);

[0205] (58).

[0206] From equations (1), (45) and (47), we can obtain:

[0207] (59);

[0208] (60).

[0209] According to assumption 1, we can directly get:

[0210] (61).

[0211] From this we can see that Theorem 1 is proved.

[0212] In order to verify the control performance of the control method proposed in this embodiment, two groups of experiments will be conducted. Figure 2 The mechanical and drive components of the tower crane test bench are described, including a suspension, a suspension rope, three drive motors and a load. In Experiment 1, the effectiveness of the proposed control method in this embodiment is evaluated by comparing it with the PD (proportional derivative) control method, the LQR (linear quadratic regulator) control method and the EEMA control method. Then, Experiment 2 shows the robustness of the proposed method to system parameter uncertainties and external disturbances.

[0213] The crane system parameters are as follows: , , , g= , , .

[0214] In addition, the dead zone related parameters are as follows: .

[0215] The desired position is set to: .

[0216] (1) Experiment 1.

[0217] In the first experiment, in order to more effectively evaluate the control performance of the proposed control method, the PD control method, LQR control method and EEMA control method were selected for comparison. The control gains of these control methods were determined by trial and error. The PD control method has the following forms:

[0218] ; (62).

[0219] in, , Represents the positioning error, and the control gain is adjusted as follows: .

[0220] The expression of LQR controller is:

[0221] ; (63).

[0222] Among them, the control gain is adjusted as follows: , .

[0223] The EEMA control method expression is:

[0224] ; (64).

[0225] Among them, the control gain is adjusted as follows: In addition, the control gain of the control method designed in this embodiment is adjusted as follows: , , , , .

[0226] Figure 3 The experimental results of four different control methods are given. It is not difficult to observe these results to find that these four control methods can successfully position the cantilever / trolley to the desired position within 4 seconds. However, the maximum load swing angle of the control method proposed in this embodiment is the smallest. In addition, Figure 3 As shown, compared with the other three control methods, the residual swing is obvious, while the residual swing of the control method proposed in this embodiment after the cantilever / trolley reaches the target position can be ignored.

[0227] (2) Experiment 2.

[0228] In this experiment, the robustness of the proposed control method will be further tested. The proposed control method is evaluated in the following three cases.

[0229] Case 1: Different load mass. The load mass is changed from 0.2 kg to 0.5 kg, keeping the control gains the same as in Experiment 1.

[0230] Case 2: Different rope lengths. The rope length is adjusted to 0.5 m, and the control gain remains the same as in Experiment 1.

[0231] Case 3: External disturbance. A disturbance of approximately 1.3 degrees is applied between 12 and 13 seconds, and the control gains remain the same as those used in Experiment 1.

[0232] from Figure 4 and Figure 5 It can be seen that the overall control performance including positioning accuracy and load swing suppression is basically unaffected by the changes in load mass and rope length. This shows that the proposed control method has good robustness to these uncertainties. Figure 6As shown in Figure 2, the proposed control method quickly offsets the introduced load swing disturbance. These results clearly demonstrate the strong robustness and reliability of the proposed control method.

[0233] Embodiment 2

[0234] The purpose of the second embodiment is to provide an adaptive neural network control system for a tower crane, comprising:

[0235] A data acquisition module, configured to: acquire tower crane system parameters and a desired position;

[0236] A control module is configured to: based on the tower crane system parameters, use an adaptive neural network as a controller to drive the auxiliary vector, the drivable error vector and the under-actuated state vector to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, based on the control input generated by the controller, combined with the control input caused by the dead zone characteristic, obtain the control input of the tower crane system, and drive the state vector of the tower crane system to track the desired position;

[0237] Among them, the driveable error vector is the error between the desired position and the driveable state vector, and the driveable state vector includes the cantilever rotation angle and the trolley displacement; the under-actuated state vector is the load swing angle; and the auxiliary vector represents the coupling relationship between the driveable and under-actuated states.

[0238] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0239] Embodiment 3

[0240] This embodiment provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor. When the program is executed by the processor, the steps in the tower crane adaptive neural network control method as described in the above-mentioned embodiment 1 are implemented.

[0241] Embodiment 4

[0242] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, the steps in the tower crane adaptive neural network control method as described in the above-mentioned embodiment 1 are implemented.

[0243] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0244] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. The tower crane adaptive neural network control method is characterized by: include: Obtain tower crane system parameters and desired position; Based on the tower crane system parameters, an adaptive neural network is used as a controller to drive the auxiliary vector, the drivable error vector and the under-actuated state vector to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, based on the control input generated by the controller and combined with the control input caused by the dead zone characteristic, the control input of the tower crane system is obtained, and the state vector of the tower crane system is driven to track the desired position; Among them, the driveable error vector is the error between the desired position and the driveable state vector, and the driveable state vector includes the cantilever rotation angle and the trolley displacement; the underactuated state vector is the load swing angle; the auxiliary vector represents the coupling relationship between the driveable and underactuated states; the auxiliary vector is estimated by the state observer, and the state observer is: ,in, is the auxiliary vector The estimate, as well as Represent the input weight vectors and estimation error The estimated value of is the observation gain matrix, , , , , is a known nominal diagonal matrix, is the basis function, express The estimate, contains a state vector that cannot be measured directly, is the system output, represents the original control input vector calculated by the control method.

2. The tower crane adaptive neural network control method according to claim 1, characterized in that: The state vector of the tower crane system includes the cantilever rotation angle, the trolley displacement and the load swing angle.

3. The tower crane adaptive neural network control method according to claim 1, characterized in that: The tower crane system parameters include the mass of the trolley and the load, the moment of inertia of the cantilever, the length of the suspension rope and the friction coefficient.

4. The tower crane adaptive neural network control method according to claim 1, characterized in that: The auxiliary vector is constructed based on the dynamic properties of the underactuated subsystem in the tower crane system, and the auxiliary vector is expressed as: ,in, is a generalized vector, for The first derivative with respect to time is for The second derivative with respect to time is The expression is = , and denote the drivable state vector and the underactuated state vector respectively, and represents a positive diagonal matrix, is the drivable error vector, , , , , , , , , , , , , , , , , , , , , , is the load mass, is the length of the suspension rope, and is the friction coefficient, g is the acceleration due to gravity, is the cantilever rotation angle, is the trolley displacement, and is the load swing angle, , , and They are , , and Abbreviation of .

5. The tower crane adaptive neural network control method according to claim 1, characterized in that: The adaptive neural network is: ,in, represents the control matrix to be adjusted, is a known nominal diagonal matrix, is the basis function, express The estimate, contains a state vector that cannot be measured directly, is the auxiliary vector, as well as Represent the input weight vector and estimation error An estimate of is obtained from the update rate.

6. The tower crane adaptive neural network control method according to claim 1, characterized in that: The tower crane system adopts a dynamic model and is decomposed into two subsystems through a full drive system approach: , ,in, and denote the drivable state vector and the underactuated state vector respectively, is the control input vector, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , and are the masses of the trolley and the load, respectively. is the cantilever moment of inertia, is the length of the suspension rope, To control the torque of the cantilever swing, is the translation control force of the trolley, , , and is the friction coefficient, g is the acceleration due to gravity, is the cantilever rotation angle, is the trolley displacement, and is the load swing angle, , , , , and They are , , , , and Abbreviation of .

7. Tower crane adaptive neural network control system, characterized in that: include: A data acquisition module, configured to: acquire tower crane system parameters and a desired position; A control module is configured to: based on the tower crane system parameters, use an adaptive neural network as a controller to drive the auxiliary vector, the drivable error vector and the under-actuated state vector to converge, so as to continuously adjust the input weight vector and the estimated value of the estimated error, so that the controller generates a control input, based on the control input generated by the controller, combined with the control input caused by the dead zone characteristic, obtain the control input of the tower crane system, and drive the state vector of the tower crane system to track the desired position; Among them, the driveable error vector is the error between the desired position and the driveable state vector, and the driveable state vector includes the cantilever rotation angle and the trolley displacement; the underactuated state vector is the load swing angle; the auxiliary vector represents the coupling relationship between the driveable and underactuated states; the auxiliary vector is estimated by the state observer, and the state observer is: ,in, is the auxiliary vector The estimate, as well as Represent the input weight vectors and estimation error The estimated value of is the observation gain matrix, , , , , is a known nominal diagonal matrix, is the basis function, express The estimate, contains a state vector that cannot be measured directly, is the system output, represents the original control input vector calculated by the control method.

8. A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor, characterized in that: When the program is executed by a processor, the steps in the tower crane adaptive neural network control method as described in any one of claims 1-6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the tower crane adaptive neural network control method as described in any one of claims 1-6 are implemented.

Images