Crane group anti-collision method based on movement trend

Abstract

The invention discloses a crane group anti-collision method based on a motion trend, which relates to the technical field of mechanical automation control, and comprises the following steps: deploying multiple sensors on a crane, collecting and integrating multi-modal data, obtaining a nonlinear state vector, constructing an improved extended Kalman filtering model based on the nonlinear state vector, and calculating the anti-collision degree of the crane group based on the improved extended Kalman filtering model. The method comprises the following steps of: deploying multiple types of sensors, performing joint optimization in combination with a deep learning algorithm, acquiring probability distribution of a motion track, distributing roles for each crane based on the probability distribution of the motion track, generating a global path for a navigator by adopting an A * algorithm, and acquiring an initial speed instruction. According to the method, the multi-source heterogeneous data are extracted and integrated into a non-linear state vector, effective fusion of the multi-source heterogeneous data is achieved, meanwhile, an improved extended Kalman filtering model is combined with a deep learning algorithm, probability distribution prediction is conducted on the motion trail of the crane, the collision risk possibly occurring in the future is effectively predicted, avoidance is conducted, and collision prevention of the crane group is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical automation control, and particularly to a method for preventing group collisions of cranes based on motion trends. Background Art

[0002] With the continuous improvement of industrial automation levels, cranes, as important logistics and manufacturing equipment, while improving work efficiency, also pose higher requirements for safety and collaborative operations. Especially in scenarios such as large ports and construction sites, the coordinated operation of multiple cranes has become the norm. In recent years, the technology of deploying sensors on cranes to achieve status monitoring and feedback control has developed rapidly, but most solutions still rely on simple rules or preset paths to avoid collisions.

[0003] Although the existing technology has made certain progress in improving the safety of cranes, there are still many challenges in dealing with the collaborative work between multiple cranes. First, traditional methods often cannot effectively integrate data from multiple sensors, resulting in an incomplete and inaccurate understanding of the crane status, which affects the quality of decision-making. Second, most existing anti-collision strategies have not fully utilized the advantages of advanced algorithms such as machine learning, especially in dealing with non-linear dynamics problems and predicting future trajectories, there are obvious deficiencies. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for preventing group collisions of cranes based on motion trends to solve the problems of insufficient data integration and insufficient ability to predict future motion trends.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for preventing group collisions of cranes based on motion trends, which includes deploying multiple sensors on the cranes, collecting and integrating multi-modal data to obtain a non-linear state vector;

[0008] Based on the non-linear state vector, an improved extended Kalman filter model is constructed and jointly optimized with a deep learning algorithm to obtain a probability distribution of motion trajectories;

[0009] Based on the probability distribution of motion trajectories, roles are assigned to each crane, and the A* algorithm is used to generate a global path for the leader to obtain an initial velocity command;

[0010] Based on the initial velocity command, the global path is updated, and the follower dynamically adjusts its speed according to the distance from the leader to generate a corrected group velocity command;

[0011] Input the corrected group velocity command into the non-linear dynamics compensation model, and generate a compensation torque by calculating the deviation between the actual position and the predicted position.

[0012] Optimize the driving signal of the hydraulic motor based on the compensation torque to obtain the final command.

[0013] The final command is published through ROS to drive the crane to run at the corrected speed, realizing collision prevention for the crane group.

[0014] As a preferred embodiment of the crane anti-collision method based on motion trend according to the present invention, wherein: the multi-sensors include an IMU, a LiDAR, and an anemometer;

[0015] The multi-modal data includes rotational angular velocity, amplitude acceleration, wind load vector, mass distribution, moment of inertia, and the distance between adjacent cranes;

[0016] The obtaining of the non-linear state vector specifically includes the following steps:

[0017] Preprocess the collected multi-modal data, eliminate low-frequency electromagnetic interference through wavelet packet decomposition, and retain high-frequency vibration characteristics;

[0018] Integrate the preprocessed multi-modal data into a non-linear state vector.

[0019] As a preferred embodiment of the crane anti-collision method based on motion trend according to the present invention, wherein: based on the non-linear state vector, construct an improved extended Kalman filter model, and perform joint optimization in combination with a deep learning algorithm to obtain the probability distribution of the motion trajectory, specifically including the following steps:

[0020] Based on the non-linear state vector, obtain the physical relationship between the state parameters in the non-linear state vector, and embed the physical relationship into the state transition equation to form a non-linear state equation;

[0021] Based on the non-linear state equation, predict the motion state of the crane at the current moment through numerical integration to obtain the predicted result of the motion state;

[0022] Use the non-linear state equation to generate the Jacobian matrix point by point, combine the current covariance matrix and the Jacobian matrix, and obtain the adjusted covariance matrix through QR decomposition;

[0023] Based on the adjusted covariance matrix, construct a Lyapunov function and obtain the optimized weight matrix through gradient descent;

[0024] Based on the optimized weight matrix, the predicted result of the motion state is corrected by the Kalman gain, and the updated covariance is used as the input for the new round of iteration to output the new predicted result of the motion state, thus completing the construction of the improved extended Kalman filter model;

[0025] The predicted result of the motion state output by the improved extended Kalman filter model is input into the LSTM neural network;

[0026] The LSTM neural network captures the long-term dependencies in the time series through the gating structure, strengthens the local feature learning ability by using residual connections, corrects the non-linear errors in the predicted result of the motion state output by the improved extended Kalman filter model, and obtains the probability distribution of the motion trajectory.

[0027] As a preferred scheme of the anti-collision method for crane groups based on motion trends according to the present invention, wherein: based on the probability distribution of the motion trajectory, roles are assigned to each crane, and the A* algorithm is used to generate a global path for the leader to obtain the initial velocity command, which specifically includes the following steps,

[0028] Extract the statistical features of the future position, velocity, and acceleration of each crane from the probability distribution of the motion trajectory;

[0029] Based on the statistical features of velocity and acceleration, the DBSCAN clustering algorithm is used to divide the cranes into leaders and followers;

[0030] Based on the position of the leader, a grid map of the crane operation environment is constructed, static obstacles and dynamic obstacles are marked, and the A* algorithm is applied to each leader for global path planning:

[0031] A weighted combination of the Euclidean distance and the Manhattan distance is used to generate a preliminary path evaluation framework;

[0032] Based on the preliminary path evaluation framework, a dynamic weight factor is introduced, and for the paths near obstacles and high-probability conflict areas, the DSF algorithm is used to generate safe paths;

[0033] The generated safe path is simplified by the Ramer-Douglas-Peucker algorithm to reduce the number of inflection points, and a smooth velocity profile is generated by cubic spline interpolation to obtain the smoothed global path;

[0034] The smoothed global path is converted into a time-position curve, and an improved PID controller is used to generate the initial velocity command.

[0035] As a preferred solution of the anti-collision method for a group of cranes based on the motion trend according to the present invention, wherein: based on the initial velocity command, update the global path, and the follower dynamically adjusts the velocity according to the distance from the leader to generate a corrected group velocity command, which specifically includes the following steps:

[0036] Based on the initial velocity command, use the A* algorithm again to update the global path;

[0037] According to the updated global path, obtain the position information of the follower and the leader through LIDAR, perform positioning by combining with the SLAM algorithm, calculate the distance between the follower and the leader, and generate a relative distance;

[0038] Obtain the error between the relative distance and the safety distance through an improved PID controller, calculate the speed adjustment amount, dynamically adjust the speed, and generate a corrected group velocity command;

[0039] The safety distance is the minimum allowable spacing defined based on the driving speed, braking ability of the crane and the position of surrounding obstacles.

[0040] As a preferred solution of the anti-collision method for a group of cranes based on the motion trend according to the present invention, wherein: input the corrected group velocity command into a non-linear dynamics compensation model, and generate a compensation torque by calculating the deviation between the actual position and the predicted position, which specifically includes the following steps:

[0041] Discretize the mass distribution and moment of inertia of the crane, and construct a non-linear dynamics compensation model including a mass matrix, an inertia tensor and a damping term through the MATLAB platform;

[0042] Convert the corrected group velocity command into a vector format through MATLAB tools and input it into the non-linear dynamics compensation model;

[0043] Through the non-linear dynamics compensation model, calculate the deviation between the actual position and the predicted position of the crane, and at the same time, based on the gradient descent and sequential quadratic programming optimization algorithms, cancel the non-linear error and generate a compensation torque.

[0044] As a preferred solution of the anti-collision method for a group of cranes based on the motion trend according to the present invention, wherein: optimize the drive signal of the hydraulic motor based on the compensation torque to obtain the final command, which specifically includes the following steps:

[0045] Based on the compensation torque, use an electro-hydraulic servo valve to control the flow rate and pressure of the hydraulic oil, and continuously adjust the position of the servo valve through a feedback loop, thereby adjusting the speed and direction of the hydraulic motor;

[0046] According to the adjusted speed and direction of the hydraulic motor, generate a drive signal for the hydraulic motor through adaptive control;

[0047] Generate the final instruction through the CAN bus based on the drive signal of the hydraulic motor.

[0048] As a preferred solution of the anti-collision method for a group of cranes based on the motion trend according to the present invention, wherein: the final instruction is published through ROS, and the crane is driven to run at the corrected speed to achieve anti-collision of the group of cranes. The specific steps are as follows.

[0049] Convert the final instruction into the Twist format and publish it through ROS to the control center of the crane that receives the motion command.

[0050] Based on the received final instruction, the crane analyzes the speed and direction information therein, and accordingly adjusts the actual motion direction and speed, and runs according to the corrected speed and direction to achieve anti-collision of the group of cranes.

[0051] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the anti-collision method for a group of cranes based on the motion trend as described in the first aspect of the present invention is implemented.

[0052] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, any step of the anti-collision method for a group of cranes based on the motion trend as described in the first aspect of the present invention is implemented.

[0053] The beneficial effects of the present invention are as follows: By deploying multiple types of sensors such as IMU and LiDAR, data including rotational angular velocity and amplitude acceleration are collected and integrated into a non-linear state vector, realizing the effective fusion of multi-source heterogeneous data. This comprehensive processing method greatly improves the cognitive accuracy of the current state of the crane, thereby enhancing the accuracy and reliability of decision-making. At the same time, the present invention adopts an improved extended Kalman filter model combined with a deep learning algorithm to predict the probability distribution of the motion trajectory of the crane. This prediction method can not only accurately capture the complex non-linear dynamic characteristics of the crane, but also effectively foresee the possible collision risks in the future, ensuring efficient and safe operation in a complex and changeable working environment. Description of the Drawings

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a flowchart of the anti-collision method for multiple cranes based on motion trend in Embodiment 1.

[0056] Figure 2 It is a flowchart of obtaining the initial velocity command through the A* algorithm in Embodiment 1. Specific Embodiments

[0057] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0058] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0059] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive with other embodiments.

[0060] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides an anti-collision method for multiple cranes based on motion trend, including the following steps:

[0061] S1. Deploy multiple sensors on the crane, collect and integrate multi-modal data, and obtain a non-linear state vector.

[0062] Specifically, the following operations are included.

[0063] Install multiple sensors at key positions of the crane to ensure that all necessary monitoring areas can be covered. For example, the IMU inertial measurement unit is usually installed on the rotating part of the crane to accurately measure the rotational angular velocity and amplitude acceleration. The LiDAR sensor is installed on the top or side of the crane to obtain the best scanning view of the surrounding environment, while the anemometer needs to be placed in the exposed external environment to ensure accurate capture of the wind load vector. Through this layout, all important physical parameters involved during the operation of the crane can be comprehensively collected, such as rotational angular velocity, amplitude acceleration, wind load vector, mass distribution, moment of inertia, and the distance between adjacent cranes.

[0064] After multi-sensors collect multi-modal data, preprocessing is carried out. Due to electromagnetic interference in the industrial environment, the original multi-modal data may contain a large amount of noise, which directly affects the accuracy of subsequent analysis. Therefore, wavelet transform is applied to the multi-modal data stream obtained from multi-sensors, and it is decomposed into multiple frequency sub-bands. According to different application scenarios and based on the statistical characteristics of the signal, a denoising threshold is set to distinguish signals from noise. The wavelet coefficients below the denoising threshold are removed to eliminate low-frequency noise, while the part above the denoising threshold is retained as the effective signal. Then, the inverse wavelet transform is used to reconstruct the signal to obtain clean multi-modal data, laying a foundation for the subsequent accurate assessment of the crane state;

[0065] The preprocessed multi-modal data is integrated by matrix operations using a neural network machine learning algorithm with the information of rotational angular velocity, amplitude acceleration, wind load vector, mass distribution, moment of inertia, and the distance between adjacent cranes, and is mapped to a 12-dimensional state space to form a non-linear state vector of 12-dimensional state parameters comprehensively describing the current state of the crane, providing multi-dimensional data for predicting the future movement trajectory of the crane and improving the prediction efficiency;

[0066] Among them, the 12-dimensional state parameters include angular velocity, acceleration, wind speed, wind direction, three-dimensional force of the hook, deformation of the boom, vibration frequency of the slewing bearing, angle of the hoisting drum and wire rope elongation, pressure distribution of the walking wheels, pressure and temperature of the hydraulic cylinder, vibration spectrum of the reducer bearing, position and force feedback of the cab operation handle.

[0067] S2. Based on the non-linear state vector, an improved extended Kalman filter model is constructed and jointly optimized with a deep learning algorithm to obtain the probability distribution of the movement trajectory.

[0068] Specifically, the following operations are included.

[0069] First, a comprehensive analysis is carried out on the operating environment of the crane and the non-linear state vector of the 12-dimensional state parameters to determine the physical relationships between the 12-dimensional state parameters. For example, the rotational angular velocity and amplitude acceleration directly affect the dynamic response of the crane boom, while the wind load vector affects the overall stability. In addition, changes in mass distribution and moment of inertia are crucial for predicting the behavior of the crane under different load conditions. The identified physical relationships are transformed into mathematical expressions, and using Lagrange's equation, these physical relationships are accurately embedded into the state transition equation to form a non-linear state equation. The above operations not only enable a more accurate description of the dynamic behavior of the crane under various working conditions but also enhance the ability to cope with uncertainties and external disturbances;

[0070] Based on the non - linear state equation, set the initial conditions according to the position, velocity and other information at the previous moment, and apply the numerical integration method to solve the non - linear state equation to obtain the motion state prediction result. In specific operations, the fourth - order Runge - Kutta method is used to solve the non - linear state equation to ensure high precision and stability;

[0071] When generating the Jacobian matrix by taking the derivative of the non - linear state equation point by point, it is first necessary to perform local linearization on the motion state vector at each time point. Through the non - linear state equation, calculate the partial derivatives of each motion state vector parameter, which can form the Jacobian matrix describing the dynamic characteristics. The motion state vector parameters include information on position, velocity, and acceleration. This process aims to capture the behavioral characteristics near the current operating point. Specifically, for the rate of change of each state variable, we need to evaluate how it is affected by other state variables. In this way, a Jacobian matrix reflecting local linearization can be constructed, which can accurately describe the interaction between state variables;

[0072] Next, combine the covariance matrix at the current moment with the Jacobian matrix to update the covariance matrix. This step is crucial because it reflects the degree of uncertainty in predicting the future motion state. To ensure the numerical stability and accuracy of the covariance matrix, the QR decomposition technique is used when combining these two matrices. QR decomposition is a method of decomposing a matrix into two simpler matrices, one of which is an orthogonal matrix and the other is an upper - triangular matrix. This method can not only simplify complex matrix operations but also effectively avoid numerical instability problems that may arise from direct calculations. By using QR decomposition, the update process of the covariance matrix can be optimized mathematically, and the calculation accuracy and stability can be maintained even in the face of high - dimensional and complex data structures. Finally, the adjusted covariance matrix is obtained after QR decomposition, which provides an important input for the subsequent algorithm;

[0073] Based on the adjusted covariance matrix, construct the Lyapunov function and obtain the optimized weight matrix through the gradient descent method. The Lyapunov function is a mathematical tool used to evaluate stability, and the smaller its value, the closer it is to the stable state. By using the gradient descent algorithm to find the weight matrix that minimizes the Lyapunov function, the stability can be effectively enhanced so that it can still maintain good performance in the face of external disturbances;

[0074] The optimized weight matrix then corrects the predicted value through the Kalman gain and uses the updated covariance as the input for the new round of iteration, and outputs the motion state prediction result, completing the construction of the improved extended Kalman filter model. This process is continuously iterated until the preset accuracy requirement is met or the convergence criterion is reached, ensuring that the improved extended Kalman filter model has high accuracy and reliability;

[0075] The motion state prediction results output by the improved extended Kalman filter model are input into the LSTM neural network. The LSTM neural network has a unique gating mechanism, including a forget gate, an input gate, and an output gate, which can effectively capture long-term dependencies in time series data. This characteristic makes the LSTM neural network very suitable for processing continuous time series data, such as the motion trajectory of a crane. In addition, in order to further improve the performance of the improved extended Kalman filter model, residual connections can be introduced into the LSTM network. This can not only strengthen the learning ability of local features but also help correct the non-linear errors in the output of the improved extended Kalman filter model. In this way, the LSTM neural network can not only refine the prediction results of the motion state but also generate a more accurate probability distribution of the motion trajectory. The formula is as follows:

[0076]

[0077] where F(x k+1 ) represents the value of the probability distribution of the motion trajectory, x k+1 represents the value of the non-linear state vector of the 12-dimensional state parameters at time point k + 1, α represents the weight coefficient with a value of 0.5, f(x k , u k ) represents the predicted value of the motion state at the next time point k + 1, x k represents the value of the non-linear state vector of the 12-dimensional state parameters at time point k, u k represents the value of the control input at time point k, y represents the value output by the LSTM neural network, and Σ represents the covariance matrix;

[0078] By obtaining the value of the probability distribution of the motion trajectory through the above formula, it provides reliable data support for subsequent tasks such as path planning. This step of joint optimization gives full play to the respective advantages of the improved extended Kalman filter and the LSTM neural network, significantly improving the overall performance and adaptability.

[0079] S3. Based on the probability distribution of the motion trajectory, assign roles to each crane, use the A* algorithm to generate a global path for the leader, and obtain the initial velocity command.

[0080] Specifically, it includes the following steps:

[0081] From the obtained probability distribution of the motion trajectory, apply feature engineering and time series analysis to extract statistical features such as the future position, speed, and acceleration of each crane, and process the extracted statistical features using the DBSCAN clustering algorithm to achieve dynamic grouping. A key advantage of the DBSCAN algorithm is its ability to identify spatial clusters of arbitrary shapes and does not require the number of clusters to be specified in advance. In actual operation, two important parameters need to be determined first: Eps (neighborhood radius) and MinPts (minimum number of points). The selection of these two parameters directly affects the final clustering effect. For a group of cranes, choosing appropriate values of Eps and MinPts can effectively group cranes with similar motion patterns into one group, thereby distinguishing the leader from the followers. Generally speaking, the leader refers to those cranes located in open areas with a clear direction of movement, while the followers are those cranes that move closely behind the leader;

[0082] After completing the role division of the cranes, the next step is to generate a global path for the leader. This process begins with constructing a grid map of the working environment, which contains not only information about static obstacles but also data on dynamic obstacles. At this stage, static obstacles such as buildings or fixed equipment can be directly added to the grid map based on the known map information. For dynamic obstacles, a real-time update mechanism may be required, such as using a sensor network to monitor other moving cranes. Subsequently, global path planning is carried out through the A* algorithm. The core of the A* algorithm is that it combines a heuristic function to evaluate the cost from the current node to the target node. The target node is preset based on the crane operation requirements, and at the same time, the actual cost of reaching the current node from the starting point is considered. This dual consideration ensures that the selected path is both the shortest and the safest. To further optimize the path, a weighted combination of Euclidean distance and Manhattan distance is used as the preliminary path evaluation framework. This method can take into account the complexity of the environment while ensuring the linearity of the path;

[0083] Based on the preliminary path evaluation framework, a dynamic weight factor is adopted to increase the cost of those paths that are closer to obstacles or have a higher collision risk. In specific operations, for each point on each candidate path, calculate its distance to the nearest obstacle, and apply the dynamic weight factor according to the distance to the nearest obstacle. For example, when a path segment is very close to an obstacle, the corresponding weight factor will increase to a higher value within the range of [1, 3], making the overall cost of these path segments increase significantly, thereby encouraging the A* algorithm to preferentially select safe paths away from obstacles;

[0084] In the case of approaching an obstacle or a high - probability conflict area, adjust the path - planning strategy. Use the DSF algorithm to deeply explore the feasible paths, especially to find the best route to bypass obstacles in a complex environment. The DSF algorithm searches the entire graph structure by recursively visiting adjacent nodes until all paths that meet the conditions are found. However, in the crane path - planning scenario, simply applying the DSF algorithm may lead to excessive unnecessary path exploration. Therefore, it is necessary to combine a dynamic weight factor to limit the search scope and give priority to path options with higher safety. Once a safe path is obtained, the Ramer - Douglas - Peucker algorithm can be used to simplify the path and reduce the number of inflection points, and then the cubic spline interpolation technique is used to generate a smooth speed profile to obtain the smoothed global path. This not only avoids the safety hazards caused by sharp turns but also improves the operation efficiency and ensures the crane runs smoothly along the predetermined path;

[0085] Convert the smoothed global path into a time - position curve and use an improved PID controller to generate the initial speed command. In this process, the PID controller is particularly crucial. It must be able to adjust the initial speed command in real - time according to the error between the current position and the target position of the crane. Therefore, the improved PID controller may include additional functions, such as the ability to adaptively adjust the proportional, integral, and differential coefficients, in order to better cope with different working conditions. This can not only improve the operation accuracy of a single crane but also enhance the collaborative working ability of the entire crane fleet, effectively reduce the collision risk, and improve the overall working efficiency.

[0086] S4. Based on the initial speed command, update the global path, and the follower dynamically adjusts its speed according to the distance from the leader to generate a corrected group speed command.

[0087] Specifically, it includes the following steps:

[0088] Use the A* algorithm to update the global path again based on the initial speed command. This process is not only to cope with environmental changes but also to optimize the path - planning to ensure that each crane can follow the safest and most effective route. Specifically, when recalculating the path each time, the A* algorithm will consider the positions and shapes of all known obstacles in the current environment and combine the latest crane position information to find an optimal path that avoids all obstacles. In addition, the A* algorithm will also update the global path according to the target position of the leader to ensure that the crane can reach the destination smoothly while preventing collisions;

[0089] According to the updated global path, use the LIDAR sensor to obtain the precise position information between the follower and the leader. The LIDAR sensor determines the distance by emitting laser beams and measuring the time it takes for the reflection to return, thus providing high-precision distance measurement data, which is crucial for accurately calculating the relative distance between the two cranes. To further improve the positioning accuracy, the SLAM (Simultaneous Localization and Mapping) algorithm is adopted. The SLAM algorithm can simultaneously perform localization and mapping in unknown and partially known environments, providing real-time and accurate position information for the cranes. By combining the distance data provided by the LIDAR sensor with the map generated by the SLAM algorithm, the exact relative distance between the follower and the leader can be accurately calculated. This dual-verification mechanism not only improves the measurement accuracy but also enhances the robustness, enabling it to operate stably in complex and changing environments;

[0090] Use an improved PID controller to handle the error between the actually measured relative distance and the predefined safety distance. The safety distance is the minimum allowable spacing comprehensively defined based on factors such as the traveling speed of the crane, braking ability, and the position of surrounding obstacles. The improved PID controller calculates the speed adjustment amount through feedback in three aspects: proportional, integral, and derivative, and adjusts the speed of the crane. First, the proportional control part directly adjusts the speed according to the current error, enabling the crane to quickly respond to any deviation in distance. The integral control part is used to eliminate long-existing errors, ensuring that the crane does not deviate from the target due to accumulated small errors. Finally, the derivative control part predicts the future error trend and makes adjustments in advance to prevent overshoot. In this way, the improved PID controller can dynamically adjust the speed of the crane to ensure that they always operate within a safe distance;

[0091] Based on the speed adjustment amount calculated by the improved PID controller, generate a corrected group speed command, which not only needs to consider the speed direction of a single crane but also the coordination requirements of the entire group. For example, when the leader changes speed, the follower needs to adjust its speed accordingly to maintain an appropriate relative distance. This dynamic adjustment mechanism enables the crane group to efficiently cooperate in a changing working environment, avoiding potential collision risks and improving the overall work efficiency. The whole process reflects a high degree of intelligence and self-adaptability, ensuring the safety and reliability of the crane group operation while achieving precise control and management of the crane group.

[0092] S5. Input the corrected group speed command into the nonlinear dynamics compensation model, and generate a compensation torque by calculating the deviation between the actual position and the predicted position.

[0093] Specifically, it includes the following operations,

[0094] To construct a non - linear dynamics compensation model that accurately describes the dynamic behavior of a crane, it is necessary to discretize the mass distribution and moment of inertia of the crane. This process involves converting the continuous mass distribution and moment of inertia into a discrete form suitable for numerical calculation. Specifically, the finite element analysis method can be used to segment the crane structure, and then a detailed mathematical model including the mass matrix, inertia tensor, and damping terms can be established on the MATLAB platform, which can accurately reflect the physical characteristics of the crane under different working conditions and provide a solid foundation for subsequent dynamic analysis;

[0095] Next, convert the corrected swarm velocity command into a format suitable for input to the non - linear dynamics compensation model. This step usually involves data format conversion to ensure that the corrected swarm velocity command can be correctly parsed by the non - linear dynamics compensation model. Using MATLAB tools, convert the corrected swarm velocity command into vector format and input it into the previously constructed non - linear dynamics compensation model. The format conversion process includes operations such as adjusting data scale and unifying units to ensure that the input data meets the requirements of the non - linear dynamics compensation model;

[0096] Use the non - linear dynamics compensation model to calculate the deviation between the actual position and the predicted position of the crane. The non - linear dynamics compensation model simulates the behavior of the crane based on the input corrected swarm velocity command and predicts its future motion trajectory. During this process, the predicted position is compared with the actually measured position to calculate the deviation between the two. To reduce the error caused by this deviation, gradient descent and sequential quadratic programming optimization algorithms are used for further correction. Gradient descent is an iterative optimization method that updates variables along the opposite direction of the gradient of the objective function to gradually approach the minimum error point, while SQP is a more advanced optimization technique that can find the optimal solution while considering constraint conditions. The combination of these two methods can effectively offset the non - linear errors caused by the simplification of the non - linear dynamics compensation model or external disturbances, thereby generating the necessary compensation torque, ensuring that even in a complex dynamic environment, the crane can operate smoothly along the predetermined path and reducing the positioning errors caused by inaccurate non - linear dynamics compensation models or external disturbances.

[0097] S6. Optimize the drive signal of the hydraulic motor based on the compensation torque to obtain the final command.

[0098] Specifically, the following operations are included,

[0099] The flow rate and pressure of hydraulic oil are controlled using an electro-hydraulic servo valve based on the compensation torque. In specific operations, the compensation torque converts the physical torque into an electrical signal through a torque sensor and transmits it to the electro-hydraulic servo valve. This servo valve can precisely adjust its opening size according to the input electrical signal, thereby regulating the oil volume and pressure entering the hydraulic chamber. This process requires not only a high-precision servo valve but also an efficient feedback loop. The feedback loop continuously monitors the oil pressure and flow rate in the hydraulic chamber and adjusts the position of the servo valve based on real-time data to ensure that the torque output by the hydraulic motor can accurately correspond to the required compensation torque. This step is the basis for achieving fine adjustment of the speed and direction of the hydraulic motor;

[0100] An adaptive control strategy is used to generate the drive signal for the hydraulic motor. Adaptive control is a technique that dynamically adjusts control parameters, including the proportional P, integral I, and derivative D gains of an improved PID controller, to cope with changing conditions. In specific operations, the adaptive control algorithm continuously monitors the actual operating state of the hydraulic motor, such as rotational speed and steering direction, and compares the data of rotational speed and steering direction with the expected values. The expected values are preset based on the target position and speed required by the task. If any deviation is detected, the improved PID controller automatically adjusts its parameters to generate a new drive signal to correct these deviations. For example, when the load of a crane suddenly increases, the adaptive control can quickly detect this change and accordingly adjust the drive signal of the hydraulic motor to ensure that the crane can maintain the preset speed and direction. In addition, the adaptive control can continuously optimize its own parameters through a learning mechanism to improve long-term stability and response speed. This method not only improves the dynamic response ability of the crane but also effectively resists external interference, ensuring that the crane can operate stably under complex working conditions;

[0101] Based on the drive signal of the hydraulic motor generated by the adaptive control, the final instruction is generated through CAN bus technology. The CAN bus is a reliable serial communication network widely used in the field of industrial automation. In specific operations, the CAN bus is used to convert the drive signal of the hydraulic motor into an instruction format recognizable by the crane control center. Each drive signal is encoded into a specific data frame and sent to each execution unit of the crane through the CAN bus. To ensure the reliability and efficiency of data transmission, the CAN bus adopts a redundant design and an error detection mechanism. For example, during the transmission process, if an error is detected in the data frame, it automatically requests retransmission to ensure that the final instruction can be accurately conveyed to the target device. This method not only simplifies the wiring and maintenance work but also improves the reliability of data transmission.

[0102] S7. The final instruction is published through ROS to drive the crane to run at the corrected speed, achieving anti-collision of the crane group.

[0103] Specifically, the following operations are included,

[0104] Extract the specific linear velocity and angular velocity from the final instruction. When extracting the velocity information, the consistency of units must be considered. For example, convert millimeters per second to meters per second to match the standard units of ROS. Convert the extracted specific linear velocity and angular velocity into a Twist-format message. The Twist-format message contains two main parts: the linear velocity represents the moving speed of the crane in the straight line direction, and the angular velocity represents the rate of change of the rotation angle. Next, construct a Twist-format message object. This object is the standard format in ROS specifically used to represent the robot's movement command. At this stage, all the linear velocity and angular velocity data are filled into the Twist-format message object to ensure that all necessary control information is ready for transmission;

[0105] Publish the linear velocity and angular velocity data in the Twist-format message object through ROS. The publishing process involves starting a ROS node. This ROS node serves as the starting point of communication and is responsible for interacting with other ROS components. When initializing the ROS node, assign a unique name to it and create a Publisher object. The Publisher needs to know the topic name to be published and the message type. Here, the Twist-format message type is used. To ensure the real-time and accuracy of the final instruction, usually check and publish new final instructions regularly in a loop, and set an appropriate message publishing frequency, such as publishing 10 times per second. This can not only ensure the timely transmission of the final instruction but also improve the response speed of the crane;

[0106] When the Twist-format message is successfully published through ROS, the crane side needs to receive and parse these messages. The crane also needs to start a ROS node and subscribe to the previously defined topic. In this way, whenever a new message arrives, the crane control center will automatically receive this data. After receiving, the crane will parse the linear velocity and angular velocity information in the Twist-format message. This process is not just a simple numerical reading but also includes the understanding and application of these values. For example, if the linear velocity value is positive, it means the crane should move forward; if it is negative, it moves backward. The angular velocity determines the steering direction and speed of the crane. According to the parsed velocity information, the motion control center of the crane generates corresponding control signals to instruct the hydraulic motor or other actuators to adjust the actual motion state of the crane. In addition, to ensure the safety and precision of the operation, implement a real-time feedback mechanism to continuously monitor the current motion state of the crane and make necessary adjustments according to the actual situation. If any deviation from the predetermined path is found, the error can be quickly corrected to ensure that the crane always moves along the correct trajectory;

[0107] Through the above operations, the entire process from calculating the final instruction to the actual execution of the crane is realized, which not only ensures the accuracy and real-time nature of the transmission of the final instruction, but also improves the response speed and precision of the crane, enabling the crane to operate efficiently and collaboratively in a complex and changing working environment and effectively avoiding the risk of collision.

[0108] This embodiment also provides a computer device applicable to the case of the anti-collision method for multiple cranes based on motion trends, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the anti-collision method for multiple cranes based on motion trends as proposed in the above embodiment.

[0109] This computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0110] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the anti-collision method for multiple cranes based on motion trends as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read Only Memory (EPROM for short), Programmable Red-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0111] In summary, the present invention realizes the effective fusion of multi-source heterogeneous data by deploying multi-type sensors such as IMUs and LiDARs, collecting data including rotational angular velocity and amplitude acceleration, and integrating them into a non-linear state vector. This comprehensive processing method greatly improves the cognitive accuracy of the current state of the crane, thereby enhancing the accuracy and reliability of decision-making. At the same time, the present invention uses an improved extended Kalman filter model combined with a deep learning algorithm to predict the probability distribution of the crane's motion trajectory. This prediction method can not only accurately capture the complex non-linear dynamic characteristics of the crane but also effectively foresee possible future collision risks, ensuring efficient and safe operation in a complex and changeable working environment.

[0112] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A crane group anti-collision method based on motion trend, characterized in that: include, Deploy multiple sensors on the crane to collect and integrate multimodal data to obtain nonlinear state vectors; Based on the nonlinear state vector, an improved extended Kalman filter model is constructed, and combined with the deep learning algorithm for joint optimization to obtain the probability distribution of motion trajectories; Based on the probability distribution of motion trajectories, a role is assigned to each crane, and the A* algorithm is used to generate a global path for the navigator to obtain the initial speed command; Based on the initial speed command, the global path is updated, and the followers dynamically adjust their speed according to the distance from the leader to generate a revised group speed command; The corrected group velocity command is input into the nonlinear dynamic compensation model, and the compensation torque is generated by calculating the deviation between the actual position and the predicted position; Based on the compensation torque, the driving signal of the hydraulic motor is optimized to obtain the final instruction; The final command is issued through ROS, driving the crane to run at the corrected speed to achieve collision avoidance for the crane group.

2. The crane group anti-collision method based on motion trend according to claim 1, characterized in that: The multi-sensor includes IMU, LiDAR and anemometer; The multimodal data includes slewing angular velocity, amplitude acceleration, wind load vector, mass distribution, moment of inertia and spacing between adjacent cranes; The nonlinear state vector is obtained by integrating multimodal data.

3. The crane group anti-collision method based on motion trend according to claim 2, characterized in that: The improved extended Kalman filter model is constructed based on the nonlinear state vector, and is combined with a deep learning algorithm for joint optimization to obtain the probability distribution of the motion trajectory, which specifically includes the following steps: The physical relationship between the state parameters in the nonlinear state vector is obtained, and the state transfer equation is embedded to form a nonlinear state equation. The motion state of the crane at the current moment is predicted by the numerical integration method to obtain the motion state prediction result; The Jacobian matrix is ​​generated by point-by-point derivation of the nonlinear state equation. The current covariance matrix and the Jacobian matrix are combined, and the adjusted covariance matrix is ​​obtained through QR decomposition. The Lyapunov function is constructed, and the optimized weight matrix is ​​obtained through gradient descent. Based on the optimized weight matrix, the motion state prediction result is corrected by the Kalman gain, and the updated covariance is used as the input of a new round of iteration to output a new motion state prediction result, thus completing the construction of the improved extended Kalman filter model. The motion state prediction results output by the improved extended Kalman filter model are input into the LSTM neural network to correct the nonlinear error and obtain the probability distribution of the motion trajectory.

4. The crane group anti-collision method based on motion trend according to claim 3 is characterized in that: The method of assigning a role to each crane based on the probability distribution of motion trajectories, using the A* algorithm to generate a global path for the navigator, and obtaining an initial speed instruction specifically includes the following steps: The statistical characteristics of crane velocity and acceleration are extracted from the probability distribution of motion trajectories, and the cranes are divided into leaders and followers using the DBSCAN clustering algorithm. Based on the position of the navigator, a grid map of the crane operating environment is constructed, static obstacles and dynamic obstacles are marked, and the A* algorithm is applied to each navigator for global path planning: A weighted combination of Euclidean distance and Manhattan distance was used to generate a preliminary path assessment framework; Based on the preliminary path evaluation framework, a dynamic weight factor is introduced to generate a safe path using the DSF algorithm for paths close to obstacles and high-probability conflict areas; The generated safe path is simplified by the Ramer-Douglas-Peucker algorithm, and a smooth velocity profile is generated using cubic spline interpolation to obtain the smoothed global path. The smoothed global path is converted into a time-position curve, and an improved PID controller is used to generate the initial speed command.

5. The crane group anti-collision method based on motion trend according to claim 4, characterized in that: Based on the initial speed instruction, the global path is updated, and the follower dynamically adjusts the speed according to the distance from the leader to generate a revised group speed instruction, which specifically includes the following steps: Based on the initial speed command, the A* algorithm is used again to update the global path; Based on the updated global path, positioning is performed through LIDAR combined with SLAM algorithm to calculate the relative distance between the follower and the leader; The error between the relative distance and the safety distance is obtained through the improved PID controller, the speed adjustment amount is calculated, the speed is adjusted dynamically, and a corrected group speed instruction is generated; The safety distance is defined based on the crane's travel speed, braking capability and the position of surrounding obstacles.

6. The crane group anti-collision method based on motion trend according to claim 5, characterized in that: The modified group speed command is input into the nonlinear dynamic compensation model, and the compensation torque is generated by calculating the deviation between the actual position and the predicted position, which specifically includes the following steps: The mass distribution and moment of inertia of the crane are discretized, and a nonlinear dynamic compensation model including mass matrix, inertia tensor and damping term is constructed through the MATLAB platform. The corrected group velocity command is input into the nonlinear dynamic compensation model to calculate the deviation between the actual position and the predicted position of the crane. At the same time, based on the gradient descent and sequential quadratic programming optimization algorithms, the nonlinear error is offset and the compensation torque is generated.

7. The crane group anti-collision method based on motion trend according to claim 6, characterized in that: The compensation torque optimizes the driving signal of the hydraulic motor and obtains the final instruction, which specifically includes the following steps: Based on the compensation torque, an electro-hydraulic servo valve is used to control the flow and pressure of the hydraulic oil, and the position of the servo valve is continuously adjusted through a feedback loop to adjust the speed and direction of the hydraulic motor; According to the adjusted speed and direction of the hydraulic motor, the driving signal of the hydraulic motor is generated through adaptive control, and the final command is generated through the CAN bus.

8. The crane group anti-collision method based on motion trend according to claim 7, characterized in that: The final instruction is issued to the control center of the crane through ROS, and after receiving the final instruction, the control center analyzes the speed and direction information therein, and adjusts the movement direction and speed to achieve collision avoidance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the crane group anti-collision method based on motion trend according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the crane group anti-collision method based on motion trend according to any one of claims 1 to 8 are implemented.

Images