Method, device and medium for cable force distribution of a rope suspension pole group tower system

Abstract

The application discloses a rope traction suspension pole holding tower system cable force distribution method, equipment and medium, and belongs to the field of power facility construction. The method comprises the following steps: S1, establishing a corresponding coordinate system according to the connection relationship of system components; S2, determining the geometric and structural parameters of the components according to the coordinate system, establishing a kinematics and dynamics model, and calculating the required six-dimensional moment of force distribution; S3, solving the initial cable force distribution by using a traditional cable force distribution algorithm according to the six-dimensional control moment, and obtaining the basic cable force; S4, constructing a simulation environment containing parameter uncertainty and rope flexibility on the dynamics model; S5, pre-training the cable force distribution strategy for incremental adjustment of the basic cable force by using the PPO algorithm for offline reinforcement learning in the simulation environment; and S6, deploying the pre-trained strategy, training the residual compensation network by using the SAC algorithm, and compensating the expected cable force of the pre-trained strategy. The method can effectively improve the outdoor operation robustness and safety of the system.

Description

Technical Field

[0001] This invention relates to the field of power facility construction technology, and in particular to a method for cable force distribution in an eight-cable traction suspended gantry hoisting device for erecting iron towers. Background Technology

[0002] With the continuous growth of electricity load, the demand for long-distance, high-capacity power transmission is also increasing. Correspondingly, the voltage levels of transmission lines are constantly being upgraded, and transmission towers play a crucial role in this process. Transmission line engineering is an important component of power grid construction; however, traditional tower construction methods suffer from low automation and high reliance on manual labor. Construction sites often require a large number of workers to perform high-altitude operations, which are complex, time-consuming, and dangerous, resulting in low overall efficiency and failing to meet the requirements of modern large-scale power grid construction.

[0003] In complex environments such as high mountains and difficult transportation, construction mainly adopts a suspended gantry tower assembly method. The suspended gantry, suspended and fixed above the tower by ropes, is a commonly used hoisting device capable of adapting to complex terrain; hereinafter referred to as the suspended gantry tower assembly system. This system mainly consists of a base frame and a slender suspended gantry. Eight retractable traction ropes are driven by a winch-motor system and connected to the upper and lower ends of the gantry. The other ends of the four ropes at the lower end of the gantry are connected to the tower frame, and the four ropes at the upper end of the gantry can also be connected to the frame or fixed to the ground. The gantry connected to the frame is called an inner suspended inner guyed gantry, and the gantry fixed to the ground is called an inner suspended outer guyed gantry. Under the coordinated action of the eight rope forces, the suspended gantry can achieve six degrees of freedom of movement in space.

[0004] Improving the mechanization and automation level of transmission line construction (especially in the tower erection stage) has become an urgent need, and the degree of mechanization has become an important indicator of the level of modern power construction. The suspended gantry tower erection system has very obvious rope traction characteristics, and its structure is very similar to that of a rope-traction parallel robot. Therefore, it is necessary to develop a rope-traction parallel robot for suspended gantry operations.

[0005] Applying the control framework of a rope-driven parallel robot to a suspended pole-mounting tower assembly mechanism, a six-dimensional control quantity is first generated using a Cartesian space controller. A force distribution algorithm is then used to convert this control quantity into the desired cable force value in the joint space, which is finally converted into a motor signal and sent to the motor for execution. Because ropes exhibit significant unidirectional force characteristics, and the conversion of control quantity into cable force requires satisfying specific static equilibrium equations, the cable force in traditional suspended pole-mounting tower assembly systems is difficult to control. Outdoor operations generally require manual monitoring, which significantly impacts construction safety. Therefore, designing an adaptive cable force distribution method with certain anti-interference capabilities and stable system tracking is a problem that needs to be solved.

[0006] In view of this, the present invention is hereby proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a method, device, and medium for cable force distribution in a rope-driven suspended gantry tower assembly system. This method optimizes the cable force level of the rope-driven suspended gantry tower assembly system based on the requirements for anti-interference capability and trajectory tracking stability during outdoor operations, while improving the overall mechanization level of the system, thereby solving the aforementioned technical problems existing in the prior art.

[0008] The objective of this invention is achieved through the following technical solution:

[0009] A method for cable force distribution in a rope-driven suspended gantry tower system includes:

[0010] Step S1: Establish a corresponding coordinate system based on the component connection relationships in the rope-traction suspended gantry tower assembly system;

[0011] Step S2: Based on the coordinate system established in Step S1, determine the geometric and structural parameters of each component and establish the kinematic and dynamic models of the rope-traction suspended pole tower system. Based on the trajectory tracking error, calculate the six-dimensional control torque required for the system to distribute force.

[0012] Step S3: Based on the six-dimensional control torque in step S2, the initial cable force distribution of the system is solved using a traditional cable force distribution algorithm to obtain the basic cable force.

[0013] Step S4: Based on the dynamic model in step S2, construct a simulation environment that includes parameter uncertainties and rope flexibility;

[0014] Step S5: Design the corresponding state space, action space and reward function in combination with the outdoor operation scenario of suspended pole; use the PPO algorithm to perform offline reinforcement learning pre-training in the simulation environment of step S4 to obtain the cable force distribution strategy for incremental adjustment of the basic cable force.

[0015] Step S6: After deploying the cable force distribution strategy of step S5, train the residual compensation network using the SAC algorithm, and design the corresponding state space, action space and reward function. Superimpose the cable force output based on the cable force distribution strategy with the cable force compensation output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system.

[0016] A processing apparatus, comprising:

[0017] At least one memory for storing one or more programs;

[0018] At least one processor is capable of executing one or more programs stored in the memory, such that when the processor executes one or more programs, the processor enables the method described in this invention.

[0019] A readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the methods described in this invention.

[0020] Compared with the prior art, the cable force distribution method, equipment and medium of the rope-traction suspended gantry tower assembly system provided by the present invention have the following advantages:

[0021] First, a coordinate system is constructed based on the assembly and connection relationship between the tower frame and the suspended gantry of the rope-traction suspended gantry tower assembly system. The system's geometric and structural parameters are calibrated, and kinematic and dynamic models are derived accordingly. An overall dynamic control architecture is then built, and a basic controller and cable force distribution algorithm are designed. Based on this, a two-stage solution strategy based on reinforcement learning—simulation pre-training and online residual learning—is proposed to effectively overcome the cable force distribution problem in suspended gantry tower assembly operations. Cable force distribution is the core execution link of dynamic control, and the solution accuracy and adaptive level directly affect the overall performance of the control system. In the pre-training stage, a high-precision numerical simulation model is constructed, introducing model uncertainty interference. Offline training is completed using the PPO algorithm, and the reward function balances disturbance rejection performance, trajectory tracking accuracy, and cable force stability. In the online operation stage, the pre-trained model is ported to a physical platform, and the residual control strategy is optimized using the SAC algorithm based on actual field conditions. This algorithm provides a new method for the dynamic control of large rope-traction mechanisms, effectively ensuring the safe and stable operation of field construction work. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating the cable force distribution method for a rope-driven suspended gantry tower assembly system provided in this embodiment of the invention.

[0024] Figure 2 This is an equivalent schematic diagram of the geometric parameters and coordinate axes of the rope-traction suspended gantry tower system provided in the embodiments of the present invention.

[0025] Figure 3 This is a schematic diagram of a commonly used rope-traction internal suspension internal pull type suspended gantry tower assembly system that can be optimized by the method of this embodiment of the invention.

[0026] Figure 4 This is a schematic diagram of a commonly used rope-traction internal suspension and external pulling type suspended gantry tower system that can be optimized by the method of this embodiment of the invention.

[0027] The component names corresponding to each mark in the diagram are: 1-Lower cable; 2-Winding mechanism; 3-Motor; 4-Suspension mast; 5-Tower frame; 6-Upper cable; 7-Rope winch-motor mechanism; 8-Traction rope; 9-Ground pulley; 10-Lifting pulley block; 11-Lifting component; 12-Control rope; 13-Suspension mast guy wire ground anchor and winch-motor mechanism; 14-Powered ground anchor and winch-motor mechanism. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them, and do not constitute a limitation on the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0029] First, the following explanations are provided for the terms that may be used in this article:

[0030] The term "and / or" means that either or both can be achieved simultaneously. For example, X and / or Y means that the case includes both "X" or "Y" as well as the three cases of "X and Y".

[0031] The terms "comprising," "including," "containing," "having," or other similar semantic descriptions should be interpreted as non-exclusive inclusion. For example, including a technical feature element (such as raw material, component, ingredient, carrier, dosage form, material, size, part, component, mechanism, device, step, process, method, reaction conditions, processing conditions, parameter, algorithm, signal, data, product or article of manufacture, etc.) should be interpreted as including not only the expressly listed technical feature element, but also other technical feature elements that are not expressly listed and are well-known in the art.

[0032] The term "composed of" excludes any technical features not expressly listed. When used in a claim, it closes the claim to exclude all technical features other than those expressly listed, except for associated conventional impurities. If the term appears only in a clause of a claim, it limits the claim to the elements expressly listed in that clause; elements recited in other clauses are not excluded from the overall claim.

[0033] Unless otherwise explicitly specified or limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this document according to the specific circumstances.

[0034] The terms “center,” “longitudinal,” “lateral,” “length,” “width,” “thickness,” “up,” “down,” “front,” “back,” “left,” “right,” “vertical,” “horizontal,” “top,” “bottom,” “inner,” “outer,” “clockwise,” and “counterclockwise” indicate the current orientation or positional relationship, and are only for the convenience and simplification of description, and do not explicitly or implicitly suggest that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this document.

[0035] The technical solution provided by this invention will be described in detail below. Contents not described in detail in the embodiments of this invention are prior art known to those skilled in the art. Where specific conditions are not specified in the embodiments of this invention, they shall be performed according to conventional conditions in the art or conditions recommended by the manufacturer. Reagents or instruments used in the embodiments of this invention whose manufacturers are not specified are all conventional products that can be purchased commercially.

[0036] like Figure 1 , Figure 2 As shown, this invention provides a method for cable force distribution in a rope-driven suspended gantry tower system, comprising the following steps:

[0037] Step S1: Establish a corresponding coordinate system based on the component connection relationships in the rope-traction suspended gantry tower assembly system;

[0038] Step S2: Based on the coordinate system established in Step S1, determine the geometric and structural parameters of each component and establish the kinematic and dynamic models of the rope-traction suspended pole tower system. Based on the trajectory tracking error, calculate the six-dimensional control torque required for the system to distribute force.

[0039] Step S3: Based on the six-dimensional control torque in step S2, the initial cable force distribution of the system is solved using a traditional cable force distribution algorithm to obtain the basic cable force.

[0040] Step S4: Based on the dynamic model in step S2, construct a simulation environment that includes parameter uncertainties and rope flexibility;

[0041] Step S5: Design the corresponding state space, action space and reward function in combination with the outdoor operation scenario of suspended pole; use the PPO algorithm to perform offline reinforcement learning pre-training in the simulation environment of step S4 to obtain the cable force distribution strategy for incremental adjustment of the basic cable force.

[0042] Step S6: After deploying the cable force distribution strategy of step S5, train the residual compensation network using the SAC algorithm, and design the corresponding state space, action space and reward function. Superimpose the cable force output based on the cable force distribution strategy with the cable force compensation output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system.

[0043] See Figure 2 , Figure 3 and Figure 4 Preferably, the main optimization object in the above method, namely the rope-traction suspended pole tower system, mainly includes: a tower frame 5 and a slender suspended pole 4, which is connected to the upper and lower ends of the suspended pole 4 by eight retractable traction ropes driven by a winch-motor mechanism 7 or 13. The four ropes connected to the lower end of the suspended mast 4 are upper tension cables 6, also called support cables, used to bear the vertical force of the suspended mast 4 and the lifting component 11. They consist of upper tension cables 6 and a rope winch-motor mechanism 7. The other end of the upper tension cable 6 is connected to the vicinity of the top of the tower frame 5. The four ropes at the upper end of the suspended mast 4 are lower tension cables 1, which are used to tighten, position and stabilize the structure. The other end of the lower tension cable 1 can be connected to the tower frame 5 and consists of lower tension cables 1 and a rope winch-motor mechanism 7. Alternatively, it can be fixed to the ground. Fixing to the ground requires the use of ground anchors. Therefore, this part consists of lower tension cables 1, suspended mast guy wires, ground anchors and winch-motor mechanism 13. The configuration connected to the tower frame 5 is called an inner suspended inner guy wire mast, while the configuration fixed to the ground is called an inner suspended outer guy wire mast. Encoders are also installed in their respective winch-motor mechanisms. Under the coordinated action of the eight rope forces, the suspended mast can complete six degrees of freedom of spatial motion.

[0044] The above-mentioned part is the structure of the part that drives the suspended mast 4 to move in the suspended mast tower assembly system, which is the object that needs to be optimized in the method of the present invention. In addition, structurally, each suspended mast is equipped with a set of lifting pulleys 10 at the top and a ground pulley 9 at the bottom of the tower frame 5. With the help of these mechanisms, the traction rope 8 can transport the lifting parts 11 through the power anchor and the power provided by the winch-motor mechanism 14 to assemble the tower frame 5 of the transmission tower. When transporting the lifting parts 11, a control rope 12 also needs to be fixed on the lifting parts 11.

[0045] This invention focuses on the design of cable force distribution for the dynamic motion of the suspended pole.

[0046] Preferably, in step S1 of the above method, establishing the corresponding coordinate system based on the component connection relationship in the rope-traction suspended gantry tower system is done in the following manner, based on the connection relationship between the tower frame and the suspended gantry of the suspended gantry tower system, establishing the world coordinate system and local coordinate system of the system, including:

[0047] Establish a world coordinate system at the center of the plane where the tower frame contacts the ground. ;

[0048] With the center of mass of the suspended mast as the origin of the body coordinate system. Establish a local coordinate system ;

[0049] When the suspended mast remains vertical, the world coordinate system and the local coordinate system remain parallel.

[0050] Preferably, in step S2 of the above method, the geometric and structural parameters of each component of the rope-traction suspended gantry tower system are determined according to the coordinate system established in step S1, including:

[0051] In the world coordinate system Below, the pose vector of the suspended boom is represented as a six-dimensional state vector. ,in This represents the position vector, where x is the position of the suspended mast on the x-axis, y is the position of the suspended mast on the y-axis, and y is the position of the suspended mast on the z-axis. Euler angle vectors representing attitude. It is the Euler angle of the precession angle. It is the Euler angle of the nutation angle. If the Euler angles are spin angles, then the local coordinate system... Relative to the world coordinate system rotation matrix Represented as:

[0052] (1);

[0053] in, (•) represents the sine function; (•) is the cosine function.

[0054] See Figure 4 In step S2 of the above method, the kinematic and dynamic models of the rope-traction suspended gantry tower system are established based on the determined geometric and structural parameters in the following manner:

[0055] In a rope-driven suspended gantry tower system, the definition is... Indicates the first on the tower frame The rope's index point, The coordinates in the world coordinate system are , They represent the first The rope index points in the world coordinate system have the x-axis, y-axis, and z-axis coordinates; definition Indicates the first The rope connection point between the root rope and the suspended pole, the rope connection point The coordinates in the local coordinate system are , They represent the first The x-coordinates, y-coordinates, and z-coordinates of the rope connection points in the local coordinate system. Indicates the first The system has eight ropes; because the rope-traction suspended gantry tower system has eight ropes, satisfy ;

[0056] Based on the vector loop equation, the relationship between the length vector of the rope and the pose of the suspended gantry in a rope-driven suspended gantry tower system is expressed as follows:

[0057] (2);

[0058] in, Indicates the first The length of the rope; This represents the direction vector of the rope, where each component... These correspond to the components along the x, y, and z axes, respectively. ; It is a rotation matrix Rope connection point in local coordinate system coordinates The dot product of represents the coordinates of the rope connection point of the suspended pole in the world coordinate system; It is a local coordinate system The origin relative to the world coordinate system The direction vector;

[0059] Taking the derivatives of both sides of equation (2) with respect to time, we can obtain the velocity of the rope. With the generalized velocity of the suspension pole The relationship between them is:

[0060] (3);

[0061] in, Indicates the hovering pole posture. It is a Jacobian matrix that satisfies:

[0062] (4);

[0063] In the above formula, the superscript T denotes the transpose of the matrix;

[0064] Generalized speed Composed of the translational velocity and angular velocity vectors of the suspended mast, the velocity used to characterize the attitude of the suspended mast is... This velocity consists of the translational velocity of the suspended mast and the Euler angular velocity vector, which is the generalized velocity. and speed The following conditions must be met:

[0065] (5);

[0066] in, It is an interaction matrix, expressed as:

[0067] (6);

[0068] (7);

[0069] In the above interaction matrix, (•) represents the sine function; (•) represents the cosine function; It is a three-dimensional identity matrix; It is a three-dimensional zero matrix; It is used to characterize the angular velocity of the suspension mast itself. With Euler angular velocity The relational Jacobian matrix is ​​the generalized velocity. and speed The main differences are in their composition;

[0070] According to the Newton-Euler formula, the dynamic model of the suspended boom can be expressed as:

[0071] (8);

[0072] In the above formula (8), and These represent the acceleration and velocity of the suspended pole posture, respectively. It is the inertial matrix of the rope-traction suspended pole-mounted tower system; It is the Coriolis matrix of the rope-traction suspended pole-mounted tower system; It is the gravity vector; It is a combination of the Jacobian matrix and the interaction matrix; It is an eight-dimensional cable force, represented by a vector, where the inertia matrix is... With Coriolis matrix The expressions are as follows:

[0073] (9);

[0074] (10);

[0075] in, It is the inertia matrix of the rope-traction suspended pole-mounted tower system; It is the mass of the suspended mast; Represents Euler angular velocity; It is an interaction matrix; express The antisymmetric matrix of a vector;

[0076] The system employs a hierarchical control approach that combines computational torque control with reinforcement learning. The torque controller calculates the six-dimensional control torque required for the force distribution of the control input based on the current state and a reference trajectory. Specifically, the trajectory tracking error of the rope-traction suspended gantry tower system is calculated based on the desired trajectory of the gantry straightening and gantry lifting segments and the current posture information. The position error, attitude error, position velocity error, and Euler angular velocity error are input into the calculation torque controller to obtain the six-dimensional control torque required for force distribution of the system. The torque controller is represented as follows:

[0077] (11);

[0078] in, It is the inertial matrix of the rope-traction suspended pole-mounted tower system; It is the expected acceleration; It is the speed error control parameter matrix; It is the pose velocity error. It is the speed of the desired trajectory; It is the position error control parameter matrix; It is a pose error. It is the expected trajectory; It is the Coriolis matrix of the rope-traction suspended pole-mounted tower system; It is the gravity vector.

[0079] Preferably, in step S3 of the above method, the traditional cable force distribution algorithm used is a zero-space projection algorithm. Based on the six-dimensional control torque from step S2, the initial cable force distribution of the system is solved using the traditional cable force distribution algorithm to obtain the basic cable force, including:

[0080] definition , For structure matrix, For a set of real numbers, the following equations must first be satisfied:

[0081] (12)

[0082] Let the first The upper and lower bounds of the rope force are respectively and Then the cable force component Must meet ;vector and Let these represent the lower and upper bounds of all cable forces, respectively. Therefore, the six-dimensional feasible generalized force set... Defined as:

[0083] (13);

[0084] in, Represents a set; It is a structure matrix The Columns, because of the structure matrix The number of columns is two more than the number of rows, so there is a two-dimensional zero space, and therefore the corresponding cable force of the foundation. The general solution is expressed as:

[0085] (14);

[0086] in, It is a particular solution of the cable force in equation (12). It is a structure matrix The false reversal; The null projection matrix satisfies , () represents the rank of the matrix, and its specific expression is:

[0087] (15);

[0088] In the above formula, It is an eight-dimensional identity matrix;

[0089] besides, As a reference cable, it needs to meet the following requirements:

[0090] (16);

[0091] To ensure the cable tension of the foundation The upper and lower limits of the cable force must be met.

[0092] Preferably, in step S4 of the above method, the dynamic model described above is a version that does not consider parameter uncertainty, so this model can be considered a nominal model. To ensure the training effect of the reinforcement learning algorithm, parameter uncertainty needs to be introduced into the simulation environment to simulate disturbances and unmodeled dynamics in the real environment. Specifically, the following types of parameter uncertainty are introduced into the simulation environment to make the pre-trained policy more robust and more in line with the needs of practical applications. Specifically, based on the dynamic model in step S2, a simulation environment containing parameter uncertainty and rope flexibility is constructed as follows:

[0093] Based on the dynamic model established above, parameter uncertainties are introduced. The corresponding dynamic model with introduced parameter uncertainties is expressed as follows:

[0094] The dynamic model that introduces parameter uncertainties is expressed as:

[0095] (17);

[0096] in, , , and This represents model parameters with uncertainties. This represents the inertia matrix of a system with uncertainties. The Coriolis matrix represents a system with uncertainty. This represents the system weight vector with uncertainty. This represents the combination of the Jacobian matrix and the interaction matrix with uncertainty, which are caused by the following factors:

[0097] (18);

[0098] in, It is the first with uncertainty The connection point of the rope is indicated. () represents the rope connection point. It is the first The degree of uncertainty at the connection points of the ropes; It is the first with uncertainty The root rope index points are represented as follows: It is the first The degree of uncertainty in the points indexed by the root rope. () represents the point where the rope index points are located; It refers to the mass of the suspended mast, which carries uncertainty. () represents the mass of the suspended mast. It refers to the degree of uncertainty in the mass of the suspended mast; It is the rotational inertia of the suspended mast with uncertainties. () represents the moment of inertia of the suspended mast. It is the degree of uncertainty in the rotational inertia of the suspended mast;

[0099] Furthermore, in practical rope drive systems, the actuator cannot instantaneously apply the desired cable force to the rope. Factors such as the elastic deformation of the rope itself, the inertia of the transmission mechanism, and the response delay of the actuator result in a dynamic lag between the actual cable force and the desired cable force in practical rope drive systems. To realistically simulate this physical process in simulations, the cable force response of each rope is modeled as a second-order low-pass dynamic system in the following manner, assuming the first... The expected tension of the rope is The actual cable force component is The rate of change of the cable force is The cable tracking dynamics model is as follows:

[0100] (19);

[0101] in, The acceleration is the change in cable force. For the first The natural frequency of the rope; The damping ratio is used; considering the different physical response characteristics of the loading phase (rope tensioning) and the unloading phase (rope relaxation), different dynamic parameters are used for the loading and unloading phases:

[0102] (20);

[0103] Among them, the inherent frequency of the loading phase Setting this to a value that enables fast response will set the inherent frequency of the unloading phase. Set to a value that simulates the slower dynamic characteristics of the rope during relaxation; damping ratio during the loading phase. Damping ratio during unloading phase The values ​​are all taken from a predetermined range at the critical damping point. For example, the damping ratio is selected as 1 during the loading stage and 1.1 during the unloading stage to avoid cable force oscillation.

[0104] Preferably, in step S5 of the above method, the corresponding state space, action space, and reward function are designed in conjunction with the outdoor operation scenario of the suspended pole, and the cable force distribution strategy is obtained by offline reinforcement learning pre-training in the simulation environment of step S4 using the PPO algorithm, including:

[0105] The lifting and hoisting task of the rope-traction suspended gantry tower assembly system is modeled as a Markov decision process:

[0106] ;

[0107] in, This represents the Markov decision-making process. The state space represents all possible states of a rope-traction suspended gantry tower system, where the system is at time [time missing]. The state is denoted as ; The action space represents the set of actions that a rope-driven, suspended gantry tower system can perform in various states; in the current state... The agent then selects actions based on the policy. Subsequently, the environment is determined based on the state transition probabilities. Evolving to the next state and give immediate rewards. ,so It is used to evaluate the completion of tasks. The discount factor is used to weigh the importance of immediate rewards against future rewards; the goal of reinforcement learning is to learn a policy. This maximizes the long-term cumulative reward that the agent obtains when continuously interacting with the environment under this strategy;

[0108] The PPO algorithm is used to generate cable force distribution strategies for offline reinforcement learning based on a simulation environment. This allows reinforcement learning to first learn the baseline distribution rules that satisfy cable force constraints and motion tracking targets. For cable force distribution problems that require a large number of samples and have high trial-and-error costs, the low sample cost and parallel data acquisition characteristics of the simulation environment can make up for the relatively low sample efficiency of the PPO algorithm with the same strategy.

[0109] State vector Constructed from perceived information and expected trajectory information, it is defined as:

[0110] (twenty one);

[0111] in, and The pose and velocity of the levitating pole at the current moment. and The corresponding expected value provided for the trajectory planner; This indicates the current actual rope force. The expected force of the previous moment;

[0112] The system employs a hierarchical control approach that combines computational torque control with reinforcement learning. The computational torque controller calculates the six-dimensional control torque of the control input based on the current state and the reference trajectory. Then we obtain the particular solution of Soli. Given reference cable force The force distribution strategy of reinforcement learning outputs actions. For reference, Soli Increment; action With reference cable force The sum is then projected using the null space matrix. Projected onto the null space of the structure matrix and superimposed on the particular solution of the cable force. Above, the desired tension is obtained:

[0113] (twenty two);

[0114] in, The action refers to the action generated by the force distribution strategy pre-trained by the PPO algorithm. During training, this action is sampled from a truncated Gaussian distribution.

[0115] (twenty three);

[0116] in, For reinforcement learning, the parameters of the Actor network; Represents a Gaussian distribution; It is the mean of this distribution. It is the standard deviation of the distribution;

[0117] By limiting the reference cable force The incremental step size ensures the continuity of cable force distribution and defines the position error. With attitude error , These are the actual location and the desired location, respectively. Let the actual pose and the desired pose be the values, then the total reward for reinforcement learning at each time step is defined as follows: for:

[0118] (twenty four);

[0119] Among them, attitude tracking accuracy reward Defined as:

[0120] (25);

[0121] Among them, the position item and posture items It employs a hybrid approach, providing both linear attraction and exponential fine-tuning capabilities: linear attraction in the large error region and exponential fine-tuning capabilities in the small error region.

[0122] (26);

[0123] In the above formula, It is the linear attraction coefficient of the positional error; It is the magnitude coefficient of the position error exponent term; It is the attenuation coefficient of the position error exponent; It is the linear attraction coefficient of attitude error; It is the magnitude coefficient of the attitude error exponent term; It is the attenuation coefficient of the attitude error exponential term;

[0124] Smooth Rewards Can guarantee the issuance of the expected cable force To maintain continuity, penalties are imposed on the first-order and second-order time differences of the desired cable force, respectively:

[0125] (27);

[0126] in, For the first A rope in Expected force and time The difference in expected tension at time, yes Time of the first The expected tension of the rope. yes Time of the first The expected tension of the rope, where 8 represents the number of ropes. This is the first A rope in Expected force and time The difference of expected force at time; the first term Suppress sudden changes in cable force commands. It is the weighting coefficient of the first-order difference penalty, the second term Suppress the oscillation mode of the cable force command. It is the weighting coefficient of the second-order difference penalty;

[0127] Soli Boundary Rewards Used to ensure that the desired cable force is within a feasible range. A second penalty is imposed on the portion that crosses the boundary, namely:

[0128] (28);

[0129] In the above formula, It is the weight coefficient of the boundary constraint penalty term; (•) is the function that takes the maximum value;

[0130] Cable force distribution uniformity reward The cable force factor (TF) is used to introduce a reward to encourage a uniform cable force distribution. The cable force factor (TF) is defined as the ratio of the minimum to the maximum expected cable force.

[0131] (29);

[0132] in, express The minimum component of the total expected cable force at any given moment; express The maximum component of the total expected cable force at any given moment. This indicates the corresponding rope number; the closer TF is to 1, the more even the tension distribution. TF is related to rewards. Defined as:

[0133] (30);

[0134] In the above formula, It is the weighting coefficient of the cable uniformity bonus item;

[0135] When the reinforcement learning agent completes the tracking of the entire trajectory with sufficient accuracy, a sparse additional reward is given. for:

[0136] (31);

[0137] in, The set completion reward value, Index of the current trajectory; This represents the total number of points on the trajectory. It is the position error threshold; It is the attitude error threshold.

[0138] For large-scale hoisting operations, if the cable force exceeds the safe range, it will have a serious impact on the equipment structure and construction safety. Therefore, it is very important to ensure that the cable force is within the safe range.

[0139] Preferably, in step S6 of the above method, after deploying the cable force distribution strategy of step S5 in the following manner, the residual compensation network is trained using the SAC algorithm, and the corresponding state space, action space, and reward function are designed. The cable force output by the cable force distribution strategy is superimposed with the cable force compensation amount output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system, including:

[0140] Learning a real-world residual compensation network using the SAC algorithm begins with designing the motion space. Since the suspended pole system has… One degree of freedom and Root-driven rope, =2 redundant ropes, select The redundant ropes form the second set of cable forces. ,the remaining The first set of cable forces is formed by the ropes. Regarding the force on the rope The rope forces and structural matrix are decomposed and represented in blocks as follows:

[0141] (32);

[0142] in, , Let the basic structure matrix and the redundant structure matrix be respectively. Substituting them into equation (12) yields:

[0143] (33);

[0144] make ,in for The solution at that time For the reason The resulting increase in cable force is expressed in the general solution as:

[0145] (34);

[0146] in, It is a structure matrix The right null space satisfies ;

[0147] Rope force sensitivity The definition is: The effect of unit cable force change in a redundant cable The maximum change in tension of the root-related rope is:

[0148] (35);

[0149] In the above formula, Represents the infinite norm;

[0150] Based on rope force sensitivity, the residual compensation network strategy selects the option with the lowest rope force sensitivity. A redundant rope is used as the motion output to ensure the stability and safety of the residual correction's impact on the remaining ropes;

[0151] During the residual learning phase, the state space Adjustments are made based on the results of the deployed cable force distribution strategy, as follows:

[0152] (36);

[0153] in, and The pose and velocity of the levitating pole at the current moment. and The corresponding expected value provided for the trajectory planner; This indicates the current actual rope force. The expected force of the previous moment;

[0154] In addition to the states introduced in offline training, the following additional states are added. The cable force distribution results calculated for the cable force distribution strategy pre-trained for the PPO algorithm;

[0155] During the residual learning phase, the action space is defined using low-dimensional rope residual motions with the lowest cable force sensitivity. The SAC residual policy network does not directly output all of the actions. Instead of adjusting the cable tension of the root rope, it combines cable tension sensitivity to generate only the correct amount. Adjustment value of the tension of the redundant rope Action definition Finally, the expected total weight was fully distributed. for:

[0156] (37);

[0157] in, The action output by the pre-trained cable force distribution strategy; The action output by the residual compensation network is mapped through the right null space of the structure matrix. Automatic transfer to all ropes, residual correction amount Located in the structure matrix Within the null space, the static equilibrium equations of the system remain unchanged; only the cable force distribution is adjusted in the null space direction. Finally, all quantities are added together to obtain the complete desired cable force. .

[0158] This design has two advantages: firstly, the motion dimension is... Down to First, it significantly reduces the search space for online exploration and improves sample efficiency; second, the initial output of the residual strategy is close to the zero vector, and the initial behavior of the system is dominated by the pre-trained strategy, ensuring a smooth transition from simulation to real-world testing.

[0159] Regarding the design of the reward function, during the training phase of the residual compensation network, since most performance requirements have already been considered in the PPO pre-training phase, the training reward function of the residual compensation network includes trajectory tracking accuracy reward, cable force smoothness reward, and expected cable force and actual cable force tracking reward for correcting the dynamic deviation between simulation and reality, as follows:

[0160] (38);

[0161] in, For the residual compensation network at each time step Training rewards; Rewards are given for trajectory tracking accuracy. It is a smooth reward; It is a reward system that tracks the expected and actual cable forces.

[0162] The attitude tracking accuracy reward for:

[0163] (39);

[0164] in, It is a position tracking error; It is attitude tracking error;

[0165] To ensure good time continuity of the cable force commands issued to the driver and to avoid mechanical shock or cable slack caused by sudden changes in commands, the cable force smoothing reward... Directly generated The punishment is as follows:

[0166] (40);

[0167] In the above formula, It is the weighting coefficient of the motion smoothing penalty term; yes The output of the time residual compensation network Each action component;

[0168] This approach suppresses abrupt changes in cable force commands between adjacent time steps, prompting the strategy to output control signals with a smooth transition. It aims to prevent excessive intervention from the residual compensation network strategy, which could affect the continuity of the pre-trained strategy and impact the performance of the suspended pole in outdoor operations.

[0169] In a real-world environment, due to unmodeled factors such as friction, cable elasticity, and transmission backlash, the actual cable force applied by the actuator varies. (Measured by cable force sensor) and the expected cable force issued by the strategy. There is often a discrepancy between the expected and actual cable force tracking rewards. To guide strategy learning and compensate for this discrepancy, a cable force tracking penalty term is introduced, whereby the expected cable force is compared with the actual cable force tracking reward. The article introduces a cable force tracking penalty term to drive the actual cable force applied. With the final expected slack The deviation between them is:

[0170] (41);

[0171] in, The first reading actually read by the cable tension sensor The actual tension of the rope; For the first batch of documents issued The expected force of the rope; This is the penalty coefficient for cable tracking accuracy. This driving strategy network gradually corrects the cable execution deviation between simulation and reality during residual learning, improving the Sim-to-Real transfer effect.

[0172] This invention also provides a processing apparatus, comprising:

[0173] At least one memory for storing one or more programs;

[0174] At least one processor is capable of executing one or more programs stored in the memory, such that when the processor executes one or more programs, the processor can implement the method of the present invention.

[0175] The present invention further provides a readable storage medium storing a computer program that, when executed by a processor, can implement the method described in the present invention.

[0176] In summary, the cable force distribution method of this invention can, on the basis of improving the overall mechanization level of the system, quickly and accurately optimize the configuration of the cable-driven suspended gantry tower system according to the anti-interference capability requirements of the cable-driven suspended gantry tower system during outdoor operations, thereby improving the quality and efficiency of power facility construction.

[0177] To more clearly demonstrate the technical solution and its effects provided by the present invention, the following detailed description of the solution provided by the embodiments of the present invention is provided with reference to specific examples.

[0178] Example 1

[0179] like Figure 1 As shown, this embodiment provides a method for cable force distribution in a rope-traction suspended gantry tower system, including the following steps:

[0180] Step S1: Establish the world coordinate system of the system based on the connection relationship between the tower frame and the suspended gantry of the rope-traction suspended gantry tower assembly system. and local coordinate system See Figure 2 ,include:

[0181] A world coordinate system is established at the center of the plane where the tower frame contacts the ground. ;

[0182] With the center of mass of the suspended mast as the origin of the body coordinate system. Establish the local coordinate system of the system ;

[0183] When the suspended mast remains vertical, the world coordinate system and the local coordinate system remain parallel.

[0184] Step S2: Determine the geometric and structural parameters of the rope-traction suspended gantry tower system based on the coordinate system established in step S1, and establish corresponding kinematic and dynamic models accordingly, including:

[0185] In the world coordinate system Below, the pose vector of the suspended boom can be expressed as: This is a six-dimensional state vector, where It is an attitude vector. Euler angle vectors representing attitude, and thus the local coordinate system Relative to the world coordinate system rotation matrix It can be represented as:

[0186] (1);

[0187] After specifying the required state variables, define Represented as the first on the tower frame The point from which the rope originates can be represented in the global coordinate system as: ,definition Represented as the first The coordinates of the rope connection point between the root rope and the suspended pole in the local coordinate system can be expressed as follows: All satisfy .

[0188] See Figure 4 According to the vector loop equation, the relationship between the length vector of the rope and the pose of the suspended pole can be expressed as:

[0189] (2);

[0190] in Indicates the first The length of the rope, The direction vector of the rope is represented by its components. These correspond to the components along the x, y, and z axes, respectively. This represents the rope connection point of the pole in the world coordinate system. Coordinate system The origin relative to the coordinate system The direction vector.

[0191] Taking the derivatives of both sides of equation (2) with respect to time, we can obtain the velocity of the rope. With the generalized velocity of the suspension pole The relationship between them is:

[0192] (3);

[0193] Among them It is a Jacobian matrix that satisfies:

[0194] (4);

[0195] Generalized speed It consists of the translational velocity and angular velocity vectors of the suspended mast, and the vector used to characterize the attitude velocity of the suspended mast is... This velocity consists of the translational velocity of the suspended mast and the Euler angular velocity vector, which is the generalized velocity. and speed The following conditions must be met:

[0196] (5)

[0197] Among them It is an interaction matrix, and its specific expression is:

[0198] (6);

[0199] (7);

[0200] In the above interaction matrix, It is a three-dimensional identity matrix; It is a three-dimensional zero matrix; It is used to characterize the angular velocity of the suspension mast itself. With Euler angular velocity The relational Jacobian matrix is ​​the generalized velocity. and speed The main differences lie in their composition.

[0201] Equation (2) above is the kinematic model of the rope-traction suspended gantry tower system. According to the Newton-Euler formula, the dynamic model of the suspended gantry in the workspace can be expressed as:

[0202] (8);

[0203] In the above formula, and These are the acceleration and velocity of the pose, respectively. It is the system's inertia matrix. It is the Coriolis matrix of the system. It is the gravity vector. It is a combination of the Jacobian matrix and the interaction matrix. For an eight-dimensional cable force, the expressions for the remaining matrices are as follows:

[0204] (9);

[0205] (10);

[0206] in, It is the system's inertia matrix. It is the mass of the suspension mast. Represents Euler angular velocity, express The antisymmetric matrix of a vector.

[0207] The system employs a hierarchical control architecture that combines computational torque control (CTC) with reinforcement learning. CTC calculates the control input, i.e., the six-dimensional control torque required for force distribution, based on the current state and reference trajectory. Specifically, based on the desired trajectory and current pose information of the rope-traction suspended gantry tower system, such as gantry straightening and gantry lifting, the trajectory tracking error of the system is calculated. The position error, attitude error, position velocity error, and Euler angular velocity error are input into the calculation torque controller to obtain the six-dimensional control torque required for force distribution of the system. The calculated torque controller here can be represented as:

[0208] (11);

[0209] in, It is the inertial matrix of the rope-traction suspended pole-mounted tower system; It is the expected acceleration; It is the speed error control parameter matrix; It is the pose velocity error. It is the speed of the desired trajectory; It is the position error control parameter matrix; It is a pose error. It is the expected trajectory; It is the Coriolis matrix of the rope-traction suspended pole-mounted tower system; It is the gravity vector.

[0210] Step S3: Based on the six-dimensional control torque obtained in step S2, the initial cable force distribution of the system is solved using a traditional cable force distribution algorithm to obtain the basic cable force. The traditional cable force distribution algorithm used in the system is a zero-space projection algorithm, defined as follows: For the structural matrix, the cable forces must first satisfy the following equations:

[0211] (12);

[0212] Let the first The upper and lower limits of the rope force are respectively and Then the cable force component Must meet .vector and Collect the lower and upper bounds of all forces. Therefore, the six-dimensional feasible generalized force set (FWS) is defined as follows:

[0213] (13);

[0214] Here It is a structure matrix The column, because Since the number of columns is two more than the number of rows, there is a two-dimensional null space. Therefore, the general solution for the cable forces of the corresponding foundation can be expressed as:

[0215] (14);

[0216] in, It is a particular solution of the cable force in equation (12). yes The pseudo-inverse of a matrix, here It is a null projection matrix that satisfies The specific expression is:

[0217] (15);

[0218] besides, It is a reference cable, and it needs to meet the following requirements:

[0219] (16);

[0220] To ensure the cable tension of the foundation The upper and lower limits of the cable force are satisfied. The above is the basic force distribution algorithm used in this invention.

[0221] Step S4: Based on the dynamic model in Step S2, construct a simulation environment incorporating parameter uncertainties and rope flexibility. The previously introduced dynamic model did not consider parameter uncertainties, so it can be considered a nominal model. To ensure the training effect of the reinforcement learning algorithm, parameter uncertainties need to be introduced into the simulation environment to simulate disturbances and unmodeled dynamics in the real environment. Specifically, the following types of parameter uncertainties are introduced into the simulation environment to make the pre-trained policy more robust and more in line with practical application requirements. The corresponding dynamic model with uncertainties can be represented as follows:

[0222] (17);

[0223] in, , , and These represent model parameters with uncertainties, which mainly consist of the following aspects:

[0224] (18);

[0225] These include the rope connection point, the rope projection point, the mass of the suspended mast, and the moment of inertia of the suspended mast; , , and These represent the degree of uncertainty of the aforementioned parameters, simulating the effects caused by changes in parameter calibration and motion. Furthermore, in actual rope-driven systems, the actuator cannot instantaneously apply the desired cable force to the rope. Factors such as the elastic deformation of the rope itself, the inertia of the transmission mechanism, and the response delay of the actuator cause a dynamic lag between the actual cable force and the desired cable force. To realistically simulate this physical process in the simulation, the cable force response of each rope is modeled as a second-order low-pass dynamic system. Let the... The expected tension of the rope is The actual cable force component is The rate of change of the cable force is The cable tracking dynamics model is as follows:

[0226] (19)

[0227] in, The acceleration is the change in cable force. For the first The natural frequency of the rope, The damping ratio is given. Considering the different physical response characteristics of the rope during the loading (tensioning) and unloading (relaxation) phases, this paper uses different dynamic parameters for the two phases:

[0228] (20);

[0229] Among them, the inherent frequency of the loading phase Set to a larger value for fast response; the inherent frequency of the unloading phase. Appropriately reduced to simulate the slower dynamic characteristics of rope relaxation; damping ratio and Values ​​near the critical damping are chosen to avoid cable force oscillations. For example, the damping ratio is selected as 1 during the loading stage and 1.1 during the unloading stage.

[0230] Step S5: Design the corresponding state space, action space, and reward function based on the outdoor operation scenario of the suspended gantry tower system. Use the PPO algorithm for offline reinforcement learning pre-training in the simulation environment of step S4 to obtain a cable force distribution strategy for incrementally adjusting the cable force on the foundation. Reinforcement learning studies how the agent maximizes long-term gains through learning strategies during its interaction with the environment. In this invention, the lifting and hoisting task of the suspended gantry tower system is modeled as a Markov decision process:

[0231] ;

[0232] in, This represents the Markov decision-making process. The state space represents all possible states of a rope-traction suspended gantry tower system, with the system at time [time missing]. The state is denoted as ; This represents the action space, that is, the set of actions that a rope-driven, suspended gantry tower system can perform in various states. In the current state... The agent then selects actions based on the policy. Subsequently, the environment is determined based on the state transition probabilities. Evolving to the next state and give immediate rewards. ,so It is used to evaluate the completion of tasks. This is a discount factor used to weigh the importance of immediate rewards against future rewards. The goal of reinforcement learning is to learn a policy. This maximizes the long-term cumulative reward that the agent obtains when continuously interacting with the environment under this strategy.

[0233] In the simulation pre-training phase, the PPO algorithm is used to generate the cable force distribution strategy, which first learns the baseline distribution rules that satisfy the cable force constraints and the motion tracking target. For cable force distribution problems that require a large number of samples and have high trial-and-error costs, the low sample cost and parallel data acquisition characteristics of the simulation environment can just make up for the relatively low sample efficiency of the PPO algorithm with the same strategy.

[0234] The state vector is constructed from perceived information and desired trajectory information, and is defined as follows:

[0235] (twenty one);

[0236] in, and The pose and velocity of the levitating pole at the current moment. and The corresponding expected value provided for the trajectory planner; Indicates the current actual cable force. This represents the desired tension command from the previous time step. This state definition helps generate time-smooth control outputs, avoiding abrupt changes between adjacent time steps.

[0237] The system employs a hierarchical control architecture that combines computational torque control (CTC) with reinforcement learning. The CTC calculates the control input based on the current state and the reference trajectory. Then we obtain the particular solution of Soli. Given reference cable force Reinforcement learning strategy output action As The increments; the sum of the two is then projected onto the null space of the structure matrix and superimposed on a particular solution of the cable force. Above, the desired tension is obtained:

[0238] (twenty two);

[0239] During training, movements Sampling from a truncated Gaussian distribution:

[0240] (twenty three);

[0241] in, These are the parameters of the Actor network. By limiting... The incremental step size ensures the continuity of cable force distribution. Position error is defined. With attitude error The total reward at each time step is defined as:

[0242] (twenty four)

[0243] Among them, attitude tracking accuracy reward Defined as:

[0244] (25);

[0245] The position and attitude terms are presented in a hybrid form, providing linear attraction in the large error region and exponential fine-tuning capability in the small error region.

[0246] (26);

[0247] in, These are the actual location and the desired location, respectively. These are the actual posture and the desired posture, respectively. The quality of trajectory tracking is a fundamental guarantee for the safety of outdoor operations.

[0248] Smooth Rewards This is to ensure the issuance of the expected cable force. For the continuity design, penalties are applied to its first-order time difference and second-order time difference respectively:

[0249] (27);

[0250] in This is a first-order difference, where 8 represents the number of ropes. The first term suppresses abrupt changes in the cable force command, and the second term suppresses the oscillation mode of the cable force command. In outdoor operating environments, severe fluctuations in the cable force command can easily induce structural resonance, causing damage to the system's mechanical structure. Therefore, maintaining the smooth continuity of the cable force command is of significant engineering importance.

[0251] Soli Boundary Rewards To ensure that the expected cable force is within a feasible range The internal design applies a second penalty to any portion that exceeds the boundary:

[0252] (28);

[0253] For large-scale hoisting operations, if the cable force exceeds the safe range, it will have a serious impact on the equipment structure and construction safety. Therefore, it is very important to ensure that the cable force is within the safe range.

[0254] Cable force distribution uniformity reward A cable force factor (TF) is used, introducing a reward to encourage a uniform cable force distribution. The cable force factor (TF) is defined as the ratio of the minimum to the maximum expected cable force.

[0255] (29);

[0256] in, express The minimum component of the total expected cable force at any given moment; express The maximum component of the total expected cable force at any given moment. This indicates the corresponding rope number; the closer TF is to 1, the more uniform the rope force distribution, which is beneficial for improving the system's dexterity and reducing the risk of overloading a single rope. Correspondingly, it enhances the anti-interference capability of large structures operating outdoors. TF-related rewards are defined as follows:

[0257] (30);

[0258] When the agent completes the tracking of the entire trajectory with sufficient accuracy, a sparse bonus is awarded. :

[0259] (31);

[0260] in, For the current trajectory index, The total number of points on the trajectory. It is the position error threshold; It is the attitude error threshold.

[0261] Step S6: After deploying the cable force distribution strategy of step S5, train the residual compensation network using the SAC algorithm, and design the corresponding state space, action space and reward function. Superimpose the cable force output based on the cable force distribution strategy with the cable force compensation output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system.

[0262] Preferably, in step S6 of the above method, the SAC algorithm is used to learn the real-machine residual compensation network. First, the motion space is designed, and then the definition of cable force sensitivity is explained. Because the suspended pole system has… One degree of freedom and Root-driven rope, =2 redundant ropes, select The redundant ropes form the second set of cable forces. ,the remaining The first set of cable forces is formed by the ropes. Regarding the force on the rope Decompose the structure. The cable forces and structural matrix are represented in blocks as follows:

[0263] (32)

[0264] in , Substituting it into equation (12), we get:

[0265] (33)

[0266] make ,in ,for The solution at that time For the reason The resulting increase in cable force. The general solution can then be expressed as:

[0267] (34)

[0268] in It is a structure matrix The right null space satisfies .

[0269] Rope force sensitivity Defined as: The effect of unit cable force change in a redundant cable Maximum tension variation of the root-related rope:

[0270] (35);

[0271] Based on the above analysis, the residual strategy with the lowest sensitivity is selected. A redundant rope is used as the motion output to ensure that the residual correction has minimal impact on the remaining ropes, thereby guaranteeing the stability and safety of the physical machine deployment.

[0272] During the residual learning phase, the state space undergoes simple adjustments based on the results of the pre-training strategy, and can be represented as:

[0273] (36);

[0274] In addition to the states introduced in offline training, a new set of states has been added. The cable force distribution results are calculated for the pre-trained PPO strategy.

[0275] In terms of motion space design, the online residual learning phase employs a low-dimensional residual motion definition consistent with cable force sensitivity analysis. Based on the aforementioned cable force distribution framework, the suspended pole system has... One degree of freedom and Root drive rope, with Redundant ropes. During the online residual learning phase, the SAC residual policy network does not directly output all... The cable tension correction for the root rope is not used; only the following parameters are generated: Adjustment value of the tension of the redundant rope Action vector is defined as Finally, the expected total weight was fully distributed. for:

[0276] (37);

[0277] in For the output of the pre-trained basic policy, For the output of the residual compensation network, this residual action is mapped through the right null space of the structure matrix. Automatic transfer to all ropes, residual correction amount Located in the structure matrix Within the null space, the static equilibrium equations of the system remain unchanged; only the cable force distribution in the null space direction is adjusted. Finally, all quantities are added together to obtain the complete desired cable force for delivery. .

[0278] This design has two advantages: firstly, the motion dimension is... Down to First, it significantly reduces the search space for online exploration and improves sample efficiency; second, the initial output of the residual strategy is close to the zero vector, and the initial behavior of the system is dominated by the pre-trained strategy, ensuring a smooth transition from simulation to real-world testing.

[0279] Next, we discuss the reward function. During the online residual policy training phase, since most performance requirements have already been considered in the PPO pre-training phase, the reward function for training the residual compensation network only considers trajectory tracking accuracy, the smoothness of the first-order cable force, and the correction of dynamic deviations between simulation and reality. At each time step... The reward signal is defined as:

[0280] (38);

[0281] in, For each time step The reward; Rewards are given for attitude tracking accuracy. It is the expected yield smoothness reward; The reward is the consistency between the expected tension and the actual tension.

[0282] The attitude tracking accuracy reward The position and orientation biases are penalized using the same hybrid linear exponential form as in the pre-training phase, as follows:

[0283] (39)

[0284] To ensure good temporal continuity of the cable force commands issued to the driver and to avoid mechanical shocks or cable slack caused by sudden changes in commands, the desired cable force smoothness bonus is provided. Directly generated The punishment is as follows:

[0285] (40);

[0286] This approach suppresses abrupt changes in cable force commands between adjacent time steps, prompting the strategy to output control signals with a smooth transition. It aims to prevent excessive intervention from the residual compensation network strategy, which could affect the continuity of the pre-trained strategy and impact the performance of the suspended pole in outdoor operations.

[0287] In a real-world environment, due to unmodeled factors such as friction, cable elasticity, and transmission backlash, the actual cable force applied by the actuator varies. (Measured by cable force sensor) and the expected cable force issued by the strategy. There is often a discrepancy between the expected and actual cable forces. To guide the strategy learning and compensate for this discrepancy, a cable force tracking penalty term is introduced, with a reward for consistency between the expected and actual cable forces. for:

[0288] (41);

[0289] in, The first reading from the actual cable force sensor The actual tension of the rope. This is the penalty coefficient. This driving strategy network gradually corrects the force execution deviation between simulation and reality during residual learning, improving the Sim-to-Real transfer effect.

[0290] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0291] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. The information disclosed in the background section of the present invention is merely intended to enhance the understanding of the overall background technology of the present invention and should not be construed as an admission or implication in any way that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for cable force distribution in a rope-traction suspended gantry tower system, characterized in that, include: Step S1: Establish a corresponding coordinate system based on the component connection relationships in the rope-traction suspended gantry tower assembly system; Step S2: Based on the coordinate system established in Step S1, determine the geometric and structural parameters of each component and establish the kinematic and dynamic models of the rope-traction suspended pole tower system. Based on the trajectory tracking error, calculate the six-dimensional control torque required for the system to distribute force. Step S3: Based on the six-dimensional control torque in step S2, the initial cable force distribution of the system is solved using a traditional cable force distribution algorithm to obtain the basic cable force. Step S4: Based on the dynamic model in step S2, construct a simulation environment that includes parameter uncertainties and rope flexibility; Step S5: Design the corresponding state space, action space and reward function in combination with the outdoor operation scenario of the suspended pole; use the PPO algorithm to perform offline reinforcement learning pre-training in the simulation environment of step S4 to obtain the cable force distribution strategy for incremental adjustment of the cable force of the foundation. Step S6: After deploying the cable force distribution strategy of step S5, train the residual compensation network using the SAC algorithm, and design the corresponding state space, action space and reward function. Superimpose the cable force output based on the cable force distribution strategy with the cable force compensation output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system.

2. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 1, characterized in that, In step S1, establishing the corresponding coordinate system based on the component connection relationships in the rope-traction suspended gantry tower system is done in the following way: The world coordinate system and local coordinate system of the system are established based on the connection relationship between the tower frame and the suspended gantry of the suspended gantry tower system, including: Establish a world coordinate system at the center of the plane where the tower frame contacts the ground. ; With the center of mass of the suspended mast as the origin of the body coordinate system. Establish a local coordinate system ; When the suspended mast remains vertical, the world coordinate system and the local coordinate system remain parallel.

3. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 2, characterized in that, In step S2, based on the coordinate system established in step S1, the geometric and structural parameters of each component of the rope-traction suspended gantry tower system are determined, including: In the world coordinate system Below, the pose vector of the suspended boom is represented as a six-dimensional state vector. ,in This represents the position vector, where x is the position of the suspended mast on the x-axis, y is the position of the suspended mast on the y-axis, and z is the position of the suspended mast on the z-axis. Euler angle vectors representing attitude. It is the Euler angle of the precession angle. It is the Euler angle of the nutation angle. If the Euler angles are spin angles, then the local coordinate system... Relative to the world coordinate system rotation matrix Represented as: (1)； in, (•) represents the sine function; (•) is the cosine function.

4. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 3, characterized in that, In step S2, the kinematic and dynamic models of the rope-traction suspended pole-mounted tower system are established based on the determined geometric and structural parameters in the following manner: In a rope-driven suspended gantry tower system, the definition is... Indicates the first on the tower frame The rope's index point, The coordinates in the world coordinate system are , They represent the first The rope index points in the world coordinate system have the x-axis, y-axis, and z-axis coordinates; definition Indicates the first The rope connection point between the root rope and the suspended pole, the rope connection point Coordinates in the local coordinate system are coordinates , They represent the first The x-coordinates, y-coordinates, and z-coordinates of the rope connection points in the local coordinate system. Indicates the first The system has eight ropes; because the rope-traction suspended gantry tower system has eight ropes, satisfy ; Based on the vector loop equation, the relationship between the length vector of the rope and the pose of the suspended gantry in a rope-driven suspended gantry tower system is expressed as follows: (2)； in, Indicates the first The length of the rope; The direction vector of the rope is represented by its components. These correspond to the components along the x-axis, y-axis, and z-axis, respectively. It is a rotation matrix Rope connection point in local coordinate system coordinates The dot product of represents the coordinates of the rope connection point of the suspended pole in the world coordinate system; It is a local coordinate system The origin relative to the world coordinate system The direction vector; Taking the derivatives of both sides of equation (2) with respect to time, we can obtain the velocity of the rope. With the generalized velocity of the suspension pole The relationship between them is: (3)； in, Indicates the hovering pole posture. It is a Jacobian matrix that satisfies: (4)； In the above formula, the superscript T denotes the transpose of the matrix; Generalized speed Composed of the translational velocity and angular velocity vectors of the suspended mast, the velocity used to characterize the attitude of the suspended mast is... This velocity consists of the translational velocity of the suspended mast and the Euler angular velocity vector, which is the generalized velocity. and speed The following conditions must be met: (5)； in, It is an interaction matrix, expressed as: (6)； (7)； In the above interaction matrix, (•) represents the sine function; (•) represents the cosine function; It is a three-dimensional identity matrix; It is a three-dimensional zero matrix; It is used to characterize the angular velocity of the suspension mast itself. With Euler angular velocity The relational Jacobian matrix is ​​the generalized velocity. and speed The main differences are in their composition; According to the Newton-Euler formula, the dynamic model of the suspended boom can be expressed as: (8)； In the above formula (8), and These represent the acceleration and velocity of the suspended pole posture, respectively. It is the inertial matrix of the rope-traction suspended pole-mounted tower system; It is the Coriolis matrix of the rope-traction suspended pole-mounted tower system; It is the gravity vector; It is a combination of the Jacobian matrix and the interaction matrix; It is an eight-dimensional cable force, in which the inertia matrix... With Coriolis matrix The expressions are as follows: (9)； (10)； in, It is the inertia matrix of the rope-traction suspended pole-mounted tower system; It is the mass of the suspended mast; Represents Euler angular velocity; It is an interaction matrix; express The antisymmetric matrix of a vector; The system employs a hierarchical control approach that combines computational torque control with reinforcement learning. The torque controller calculates the six-dimensional control torque required for the force distribution of the control input based on the current state and a reference trajectory. Specifically, the trajectory tracking error of the rope-traction suspended gantry tower system is calculated based on the desired trajectory of the gantry straightening and gantry lifting segments and the current pose information. The position error, attitude error, position velocity error, and Euler angular velocity error are input into the calculation torque controller to obtain the six-dimensional control torque required for force distribution of the system. The torque controller is represented as follows: （11）； in, It is the inertial matrix of the rope-traction suspended pole-mounted tower system; It is the expected acceleration; It is the speed error control parameter matrix; It is the pose velocity error. It is the speed of the desired trajectory; It is the position error control parameter matrix; It is a pose error. It is the expected trajectory; It is the Coriolis matrix of the rope-traction suspended pole-mounted tower system; It is the gravity vector.

5. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 4, characterized in that, In step S3, the traditional cable force distribution algorithm used is a zero-space projection algorithm. Based on the six-dimensional control torque from step S2, the initial cable force distribution of the system is solved using the traditional cable force distribution algorithm to obtain the basic cable forces, including: definition , For structure matrix, For a set of real numbers, the following equations must first be satisfied: （12） Let the first The upper and lower bounds of the rope force are respectively and Then the cable force component Must meet ;vector and Let these represent the lower and upper bounds of all cable forces, respectively. Therefore, the six-dimensional feasible generalized force set... Defined as: （13）； in, Represents the elements in the FWS set; It is a structure matrix The Columns, because of the structure matrix The number of columns is two more than the number of rows, so there is a two-dimensional zero space, and therefore the corresponding basic cable force... The general solution is expressed as: （14）； in, It is a particular solution of the cable force in equation (12). It is a structure matrix The false reversal; It is a null projection matrix that satisfies , () denotes the rank of the matrix. The specific expression is: （15）； In the above formula, It is an eight-dimensional identity matrix; besides, As a reference cable, it needs to meet the following requirements: （16）； To ensure the cable tension of the foundation The upper and lower limits of the cable force must be met.

6. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 5, characterized in that, In step S4, based on the dynamic model of step S2, a simulation environment containing parameter uncertainties and rope flexibility is constructed in the following manner: The dynamic model that introduces parameter uncertainties is expressed as: （17）； in, , , and This represents model parameters with uncertainties. This represents the inertia matrix of a system with uncertainty. The Coriolis matrix represents a system with uncertainty. This represents the system weight vector with uncertainty. This represents the combination of the Jacobian matrix and the interaction matrix with uncertainties, which are caused by the following factors: （18）； in, It is the first with uncertainty The connection point of the rope is indicated. () represents the rope connection point. It is the first The degree of uncertainty at the connection points of the ropes; It is the first with uncertainty The root rope index points are represented as follows: It is the first The degree of uncertainty in the points indexed by the root rope. () represents the point where the rope index points are located; It refers to the mass of the suspended mast, which carries uncertainty. () represents the mass of the suspended mast. It refers to the degree of uncertainty in the mass of the suspended mast; It is the rotational inertia of the suspended mast with uncertainties. () represents the moment of inertia of the suspended mast. It is the degree of uncertainty in the rotational inertia of the suspended mast; In practical rope-driven systems, the actual cable force exhibits a dynamic lag relative to the desired cable force. The cable force response of each rope is modeled as a second-order low-pass dynamic system as follows: Let the... The expected tension of the rope is The actual cable force component is The rate of change of the cable force is The cable tracking dynamics model is as follows: （19）； in, The acceleration is the change in cable force. For the first The natural frequency of the rope; The damping ratio is used; considering the different physical response characteristics of the loading phase (rope tensioning) and the unloading phase (rope relaxation), different dynamic parameters are used for the loading and unloading phases: （20）； Among them, the inherent frequency of the loading phase Setting this to a value that enables fast response will set the inherent frequency of the unloading phase. Set to a value that simulates the slower dynamic characteristics of the rope during relaxation; damping ratio during the loading phase. Damping ratio during unloading phase All values ​​are taken within a predetermined range at the critical damping point to avoid cable force oscillation.

7. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 6, characterized in that, In step S5, the corresponding state space, action space, and reward function are designed in accordance with the outdoor operation scenario of the suspended pole, and the cable force distribution strategy is obtained by offline reinforcement learning pre-training in the simulation environment of step S4 using the PPO algorithm, including: The lifting and hoisting task of the rope-traction suspended gantry tower assembly system is modeled as a Markov decision process: ； in, This represents the Markov decision-making process. The state space represents all possible states of a rope-traction suspended gantry tower system, where the system is at time [time missing]. The state is denoted as ; The action space represents the set of actions that a rope-driven, suspended gantry tower system can perform in various states; in the current state... The agent then selects actions based on the policy. Subsequently, the environment is determined based on the state transition probabilities. Evolving to the next state and give immediate rewards. ,so It is used to evaluate the completion of tasks. The discount factor is used to weigh the importance of immediate rewards against future rewards; the goal of reinforcement learning is to learn a policy. This maximizes the long-term cumulative reward that the agent obtains when continuously interacting with the environment under this strategy; The PPO algorithm is used to perform simulation pre-training on offline reinforcement learning in a simulation environment to generate cable force distribution strategies, so that reinforcement learning first learns the baseline distribution rules that satisfy cable force constraints and motion tracking targets; State vector Constructed from perceived information and expected trajectory information, it is defined as: （21）； in, and The pose and velocity of the levitating pole at the current moment. and The corresponding expected value provided for the trajectory planner; This indicates the current actual rope force. The expected force of the previous moment; The torque controller calculates the six-dimensional control torque of the control input based on the current state and the reference trajectory. Then we obtain the particular solution of Soli. Given reference cable force The force distribution strategy of reinforcement learning outputs action vectors. For reference, Soli Increment; action vector With reference cable force The sum is then projected using the null space matrix. Projected onto the null space of the structure matrix and superimposed on the particular solution of the cable force. Above, the desired tension is obtained: （22）； in, The action refers to the action generated by the force distribution strategy pre-trained by the PPO algorithm. During training, this action is sampled from a truncated Gaussian distribution. （23）； in, For reinforcement learning of Actor network parameters; Represents a Gaussian distribution; It is the mean of this distribution. It is the standard deviation of the distribution; By limiting the reference cable force The incremental step size ensures the continuity of cable force distribution and defines the position error. With attitude error , These are the actual location and the desired location, respectively. Let the actual pose and the desired pose be the values, then the total reward for reinforcement learning at each time step is defined as follows: for: （24）； Among them, attitude tracking accuracy reward Defined as: （25）； Among them, the position item and posture items It employs a hybrid approach, providing both linear appeal and exponential fine-tuning capabilities: （26）； In the above formula, It is the linear attraction coefficient of the positional error; It is the magnitude coefficient of the position error exponent term; It is the attenuation coefficient of the position error exponent; It is the linear attraction coefficient of attitude error; It is the magnitude coefficient of the attitude error exponent term; It is the attenuation coefficient of the attitude error exponential term; Smooth Rewards Can guarantee the issuance of the expected cable force To maintain continuity, penalties are imposed on the first-order and second-order time differences of the desired cable force, respectively: （27）； in, For the first A rope in Expected force and time The difference in expected tension at time, yes Time of the first The expected tension of the rope. yes Time of the first The expected tension of the rope, where 8 represents the number of ropes. This is the first A rope in Expected force and time The difference of expected force at time; the first term Suppress sudden changes in cable force commands. It is the weighting coefficient of the first-order difference penalty, the second term Suppress the oscillation mode of the cable force command. These are the weighting coefficients of the second-order difference penalty; Soli Boundary Rewards Used to ensure that the desired cable force is within a feasible range. A second penalty is imposed on the portion that crosses the boundary, namely: （28）； In the above formula, It is the weight coefficient of the boundary constraint penalty term; (•) is the function that takes the maximum value; Cable force distribution uniformity reward The cable force factor (TF) is used to introduce a reward to encourage a uniform cable force distribution. The cable force factor (TF) is defined as the ratio of the minimum to the maximum expected cable force. （29）； in, express The minimum component of the total expected cable force at any given moment; express The maximum component of the total expected cable force at any given moment. This indicates the corresponding rope number; the closer TF is to 1, the more even the tension distribution. TF is related to rewards. Defined as: （30）； In the above formula, It is the weighting coefficient of the cable uniformity bonus item; When the reinforcement learning agent completes the tracking of the entire trajectory with sufficient accuracy, a sparse additional reward is given. for: （31）； in, The set completion reward value, Index of the current trajectory; This represents the total number of points on the trajectory. It is the position error threshold; It is the attitude error threshold.

8. The cable force distribution method for the rope-traction suspended gantry tower system according to claim 7, characterized in that, In step S6, after deploying the cable force distribution strategy from step S5 as follows, the residual compensation network is trained using the SAC algorithm, and the corresponding state space, action space, and reward function are designed. The cable force output by the cable force distribution strategy is superimposed with the cable force compensation amount output by the residual compensation network to obtain the corrected expected cable force used to drive the rope-traction suspended gantry tower system, including: Learning a real-world residual compensation network using the SAC algorithm begins with designing the motion space. Since the suspended pole system has… One degree of freedom and Root drive rope, =2 redundant ropes, select The redundant ropes form the second set of cable forces. ,the remaining The first set of cable forces is formed by the ropes. Regarding the force on the rope The rope forces and structural matrix are decomposed and represented in blocks as follows: （32）； in, , Let the basic structure matrix and the redundant structure matrix be respectively. Substituting them into equation (12) yields: （33）； make ,in for The solution at that time For the reason The resulting increase in cable force is expressed in the general solution as: （34）； in, It is a two-dimensional identity matrix. It is a structure matrix The right null space satisfies ; Rope force sensitivity The definition is: The effect of unit cable force change in a redundant cable The maximum change in tension of the root-related rope is: （35）； In the above formula, Represents the infinite norm; Based on rope force sensitivity, the residual compensation network strategy selects the option with the lowest rope force sensitivity. A redundant rope is used as the motion output to ensure the stability and safety of the residual correction's impact on the remaining ropes; During the residual learning phase, the state space Adjustments are made based on the results of the deployed cable force distribution strategy, as follows: （36）； in, and The pose and velocity of the levitating pole at the current moment. and The corresponding expected value provided for the trajectory planner; This indicates the current actual rope force. The expected force of the previous moment; In addition to the states introduced in offline training, the following additional states are added. The cable force distribution results calculated for the cable force distribution strategy pre-trained for the PPO algorithm; During the residual learning phase, the action space is defined using low-dimensional rope residual motions with the lowest cable force sensitivity. The SAC residual policy network does not directly output all of the actions. Instead of adjusting the cable tension, it combines cable tension sensitivity to generate only the cable tension correction amount. Adjustment value of the tension of the redundant rope Action vector is defined as Finally, the expected total weight was fully distributed. for: （37）； in, It is the null projection matrix; The action vector output by the pre-trained force distribution strategy; This is the action vector output by the residual compensation network, which is mapped through the right null space of the structure matrix. Automatic transfer to all ropes, residual correction amount Located in the structure matrix Within the null space, the static equilibrium equations of the system remain unchanged; only the cable force distribution is adjusted in the null space direction. Finally, all quantities are added together to obtain the complete desired cable force. ; During the training phase of the residual compensation network, the training reward function includes a trajectory tracking accuracy reward, a cable force smoothness reward, and a reward for tracking the expected and actual cable forces to correct for dynamic deviations between simulation and reality. These are: （38）； in, For the residual compensation network at each time step Training rewards; Rewards are given for trajectory tracking accuracy. It is a smooth reward; It is a reward system that tracks the expected and actual cable forces. The attitude tracking accuracy reward for: （39）； in, It is a position tracking error; It is attitude tracking error; The cable smoothing reward Directly generated The punishment is as follows: （40）； In the above formula, It is the weighting coefficient of the motion smoothing penalty term; yes The output of the time residual compensation network Each action component; The expected tension and actual tension tracking reward The article introduces a cable force tracking penalty term to drive the actual cable force applied. With the final expected slack The deviation between them is: （41）； in, The first reading actually read by the cable tension sensor The actual tension of the rope; For the first The expected force of the rope; This is the penalty coefficient for cable tracking accuracy.

9. A processing device, characterized in that, include: At least one memory for storing one or more programs; At least one processor is capable of executing one or more programs stored in the memory, such that when the one or more programs are executed by the processor, the processor can perform the method according to any one of claims 1-8.

10. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it can implement the method described in any one of claims 1-8.

Images