Tower crane control method, system and terminal device based on assisted driving

Abstract

An embodiment of the present application provides a tower crane control method, system, and terminal device based on assisted driving, belonging to the technical field of tower crane operation control. The method includes: deploying a control system in the tower crane device; constructing an adaptive decision-making model of the tower crane device by using the A3C algorithm according to the historical status information, historical mode selection information, and historical execution situation of the tower crane operation task of the tower crane device, so as to select the assisted driving mode most matching the tower crane device; obtaining the real-time status information of the tower crane device, and configuring a target mode matching the real-time status information by using the adaptive decision-making model; receiving an operation instruction sent by the user to the tower crane device, and constructing a tower crane task model of the tower crane device in the target mode based on the operation instruction; displaying the tower crane task model to the user; after the user confirms the tower crane task model, sending the tower crane task model to the control system to control the tower crane device to complete the corresponding operation task according to the operation time sequence.

Description

Technical Field

[0001] This application relates to the technical field of tower crane operation control, and in particular to a tower crane control method, system, and terminal device based on assisted driving. Background Art

[0002] The traditional tower crane control method mainly involves manual operation by the tower crane driver in the cab.

[0003] The tower crane driver operates control devices such as joysticks and buttons in the cab to achieve actions such as luffing, hoisting, and slewing. These devices are connected to actuators such as motors and hydraulic systems, and the driver manually adjusts the tower crane according to the lifting requirements. During the operation, the driver needs to judge the lifting parameters and control the tower crane based on the instructions such as lifting and lowering the hook conveyed by the ground commander through gestures, walkie-talkies, etc. This control method requires high operation skills and experience of the driver. In complex working conditions such as multi-tower crane coordination or narrow spaces, it is easy to cause operation errors and safety accidents due to the lack of driver experience. At the same time, due to the limited vision of the driver, it may be difficult to accurately judge the lifting position and posture for some complex lifting tasks, affecting work efficiency. Summary of the Invention

[0004] The main purpose of the embodiments of this application is to provide a tower crane control method, system, and terminal device based on assisted driving, aiming to solve at least one technical problem existing in the traditional tower crane control method.

[0005] In a first aspect, the embodiments of this application provide a tower crane control method based on assisted driving, including:

[0006] Deploy a control system in the tower crane equipment; the control system is connected to the linkage console, and multiple auxiliary driving modes are integrated in the control system; the multiple auxiliary driving modes are respectively used to assist the user to control the tower crane equipment by adopting corresponding operation modes; the multiple auxiliary driving modes at least include: linkage console mode, voice driving mode, fixed-distance driving mode, macro-control mode, body-sensing control mode; according to the historical state information of the tower crane equipment, historical mode selection information, and historical execution situation of the tower crane operation task, use the A3C algorithm to construct an adaptive decision-making model of the tower crane equipment to select the auxiliary driving mode that best matches the tower crane equipment; obtain the real-time state information of the tower crane equipment, and use the adaptive decision-making model to configure the target mode that matches the real-time state information; receive the operation instructions issued by the user to the tower crane equipment, and construct a tower crane task model of the tower crane equipment in the target mode based on the operation instructions; multiple task nodes are sequentially arranged in the tower crane task model in the order of operation time sequence from front to back, and each task node corresponds to a to-be-executed operation task; the operation instructions are at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a body-sensing control instruction; display the tower crane task model to the user; after the user confirms the tower crane task model, send the tower crane task model to the control system to control the tower crane equipment to complete the corresponding operation task according to the operation time sequence.

[0007] In a second aspect, an embodiment of the present application provides a tower crane control system based on auxiliary driving, including:

[0008] A deployment module for deploying a control system in a tower crane device; the control system is connected to a linkage console, and multiple auxiliary driving modes are integrated in the control system; the multiple auxiliary driving modes are respectively used to assist the user in implementing control operations on the tower crane device in corresponding operation modes; the multiple auxiliary driving modes at least include: a linkage console mode, a voice driving mode, a fixed-distance driving mode, a macro-control mode, and a somatosensory control mode; a decision-making module for constructing an adaptive decision-making model of the tower crane device by using the A3C algorithm according to the historical state information of the tower crane device, the historical mode selection information, and the historical execution situation of the tower crane operation task, so as to select the auxiliary driving mode most matching the tower crane device; a configuration module for obtaining the real-time state information of the tower crane device and configuring a target mode matching the real-time state information by using the adaptive decision-making model; a task module for receiving an operation instruction issued by the user to the tower crane device and constructing a tower crane task model of the tower crane device in the target mode based on the operation instruction; multiple task nodes are sequentially arranged in the tower crane task model in the order from first to last according to the operation time sequence, and each task node corresponds to a to-be-executed operation task; the operation instruction is at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a somatosensory control instruction; a display module for displaying the tower crane task model to the user; a sending module for, after the user confirms the tower crane task model, sending the tower crane task model to the control system to control the tower crane device to complete the corresponding operation task according to the operation time sequence.

[0009] In a third aspect, an embodiment of the present application further provides a terminal device, which includes a processor and a memory for storing a computer program; the processor is used to execute the computer program and implement the tower crane control method based on auxiliary driving described in the first aspect or any embodiment of the present application when executing the computer program.

[0010] The embodiments of this application provide a tower crane control method, system, and terminal device based on assisted driving. The method includes: First, deploy a control system connected to the linkage console in the tower crane device, integrating multiple assisted driving modes such as the linkage console, voice, fixed distance, macro distance, and body sensing to meet the needs of different working conditions and operators. For example, those with insufficient experience can select the fixed-distance mode to reduce the operation difficulty and facilitate the operator to flexibly switch modes. The A3C algorithm is used to construct an adaptive decision-making model based on the historical state of the tower crane device, mode selection, and execution of operation tasks, automatically recommending the most suitable assisted driving mode to assist in reducing mode selection errors, and the model is optimized with new data. Obtain the real-time status information of the tower crane, and the adaptive decision-making model quickly configures the matching target mode. Under complex working conditions, adjust the mode according to the real-time distance, obstacle position, etc. to avoid safety accidents and ensure the stable operation of the tower crane. Finally, receive the user operation instruction, construct the tower crane task model in the target mode and display it, and issue it to the control system after the user confirms. The task model sets multiple task nodes according to the operation time sequence, providing clear operation guidelines to assist the operator in understanding the task process and requirements in advance, reducing operation delays, and improving efficiency. Issuing after confirmation can increase the field of vision and improve the operation safety, accuracy, and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic flowchart of a tower crane control method based on assisted driving provided by the embodiments of this application;

[0012] Figure 2 It is a schematic block diagram of the module structure of a tower crane control system based on assisted driving provided by the embodiments of this application;

[0013] Figure 3 It is a schematic block diagram of the structure of a terminal device provided by the embodiments of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] In view of the technical problems existing in the traditional tower crane control method, the embodiments of the present application propose a tower crane control method, system and terminal device based on assisted driving. Specifically, aiming at the problems such as high requirements for drivers' operation skills, easy mistakes in complex working conditions, and limited vision affecting efficiency, the embodiments of the present application achieve efficient and safe operation through multi-mode integration and intelligent decision-making. The control system integrates multiple assisted driving modes such as a linkage console, voice, fixed distance, macro distance, and somatosensory. The driver can flexibly select according to the task requirements. For example, the fixed-distance mode can be used for simple lifting to reduce the operation difficulty. At the same time, the adaptive decision-making model based on the A3C algorithm can automatically match the optimal mode according to the historical information of the tower crane, assisting drivers with insufficient experience to reduce mistakes. In complex working conditions, the system can obtain the device status information in real time and dynamically select the appropriate mode. For example, when an obstacle is detected, it switches to the macro-control mode to avoid the collision risk, and plans the operation time sequence by constructing a tower crane task model to ensure the orderly progress of the operation. In addition, it supports multiple operation instructions. The somatosensory control mode combined with a high-definition camera and AR device broadens the driver's vision. Together with the visualization display and confirmation mechanism of the tower crane task model, it helps the driver master the task process in advance, improving the operation efficiency and ensuring the operation safety.

[0015] The embodiments of the present application provide a tower crane control method, system and terminal device based on assisted driving. Among them, the tower crane control method based on assisted driving can be applied to a terminal device. The terminal device can be a linkage console in the tower crane device, or a linkage console in the tower crane device and a mobile terminal communicating with the linkage console, such as electronic devices like mobile phones, virtual reality devices, tablet computers, laptop computers, desktop computers, wearable devices, etc. The terminal device can be a server connected to the tower crane device, or a server cluster. The above connection methods can be realized through a hardware circuit or through a communication module.

[0016] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other. Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a tower crane control method based on assisted driving provided by the embodiments of the present application.

[0017] As Figure 1 shown, the tower crane control method based on assisted driving includes steps S101 to S106.

[0018] Step S101: Deploy a control system in the tower crane device.

[0019] In the embodiments of the present application, the control system is the core part of the intelligent operation of the tower crane. The control system realizes the comprehensive management and control of the tower crane through hardware devices (such as sensors, controllers, communication modules, etc.) and software programs. For example, in terms of hardware, the sensors collect the operation data of the tower crane in real time, such as the position of the hook, the load weight, the angle of the boom, etc. The controller analyzes and processes these data. The communication module is responsible for data transmission with other devices (such as the linkage console, the remote monitoring center, etc.). The software program includes various control algorithms, decision-making logics, operation interfaces, etc., and can realize the real-time monitoring, fault diagnosis, safety protection, and intelligent control of the operation state of the tower crane.

[0020] Exemplarily, the control system is connected to the linkage console, and multiple assisted driving modes are integrated in the control system. The control system is connected to the linkage console, which not only supports the traditional manual operation mode but also integrates multiple advanced assisted driving modes, enabling the tower crane to have more flexible and intelligent operation capabilities.

[0021] The linkage console is the core hardware device for operating the tower crane. It can be understood as the interaction hub between the tower crane driver and the mechanical actions of the tower crane, and it is the core component that can convert the driver's operation intention into the control instructions for the actual operation of the tower crane. Exemplarily, the linkage console can be installed in the tower crane cab and is composed of multiple joysticks, buttons, switches, indicator lights, somatosensory control components, multiple sensors or data acquisition modules matching the assisted driving modes, etc.

[0022] In the embodiments of the present application, multiple assisted driving modes are respectively used to assist the user in implementing the control operation of the tower crane equipment by adopting the corresponding operation modes. The aim is to assist the user in more efficiently and safely operating the tower crane from different dimensions, so as to comprehensively improve the intelligent level and safety of tower crane operation.

[0023] Exemplarily, the multiple assisted driving modes at least include: the joystick mode, the voice driving mode, the distance-fixed driving mode, the macro-control mode, and the somatosensory control mode. Each mode is designed for different operation scenarios and requirements. In the joystick mode, with the joystick handle and buttons as the operation core, instructions such as the lifting and lowering of the boom and the movement of the hook are accurately transmitted to the control system, retaining the characteristics of traditional operations, being easy to get started, standardizing the process, and being applicable to common operation scenarios. In the voice driving mode, instructions such as "lift the hook by 5 meters" are converted into text through the voice recognition module and executed after parsing. Combined with voiceprint recognition to verify the identity, it ensures operation safety. This mode frees the hands and is especially suitable for scenarios where the hands are busy or the space is limited. In the distance-fixed driving mode, after inputting the target distance, the system automatically calculates the hook movement parameters and uses lidar or visual sensors to calibrate in real time to accurately control the hook position, which is applicable to operations with high precision requirements such as accurately placing building components. In the macro-control mode, instructions are input through a knob or touch screen to achieve centimeter-level fine movement of the hook, and the position information is fed back in real time to meet the requirements of close-range high-precision installation operations and reduce operation risks. In the somatosensory control mode, the high-definition camera or AR device under the tower is used to collect images, and the operator directly operates the tower crane through the touch screen or AR glasses, breaking the communication barrier between the tower top and the tower bottom, providing a broad vision, and significantly improving the operation efficiency and safety in complex environments. It can be understood that in the related art, the traditional joystick is mainly used to control the basic movements of the tower crane, such as the lifting and lowering of the boom, the movement of the hook, and the slewing. In the embodiments of the present application, to cooperate with the multiple assisted driving modes, the joystick integrates more functions. For example, in terms of cooperation with the voice driving mode, the joystick has newly added a voice instruction receiving and parsing module, which can convert the instructions parsed by the voice recognition module into corresponding control signals to achieve precise control of the tower crane. When receiving and recognizing the voice instruction "lift the hook by 5 meters", the joystick quickly converts this instruction into a signal to control the operation of the hook lifting motor, making the hook accurately rise by 5 meters. For the distance-fixed driving mode, the joystick integrates a target distance input interface and function modules related to the automatic control algorithm. After the operator inputs the target distance, the joystick can automatically adjust the hook movement speed and distance according to the built-in algorithm to ensure that the tower crane accurately reaches the specified position.

[0024] In terms of optimizing the operation convenience, the joystick reorganizes the key components such as the handle and buttons, arranges them according to the operation frequency and relevance, reduces the operation range and switching time. At the same time, it is centrally arranged according to the operation logic, and realizes linkage control in combination with multiple assisted driving modes, simplifying the complex process with one-key buttons. The operation components are set with obvious markings and tactile distinctions to improve the operation accuracy and efficiency in poor light or emergency situations.

[0025] At the security protection level, the operation authority management function is added. By integrating with biometric recognition technologies such as voiceprint recognition and fingerprint recognition, only authorized operators can operate the control console, preventing unauthorized personnel from causing safety accidents due to misoperation. Redundant design and fault isolation technology are adopted inside the control console. When a certain circuit module fails, it will not affect the normal operation of other modules, ensuring the continuity and stability of tower crane operation.

[0026] To better integrate with other auxiliary driving modes and achieve intelligent interaction, the control console is equipped with an intelligent display screen. This display screen can real-time display information such as the operating status of the tower crane, the execution status of operation instructions, and the working parameters of various auxiliary driving modes. In the macro-control mode, the display screen can accurately display the centimeter-level moving position of the hook, providing accurate feedback for the operator. The control console has communication interfaces with other intelligent devices or systems, enabling remote monitoring and operation. Managers can connect to the control console through terminal devices such as mobile phones and computers, real-time understand the operating conditions of the tower crane, and remotely operate the tower crane when necessary, improving management efficiency and emergency response capabilities. In the multi-tower operation scenario, the control console of this tower crane can interact with the control consoles of other tower cranes, plan the operating paths of tower cranes through intelligent algorithms, and avoid collision accidents.

[0027] Step S102: According to the historical status information of the tower crane equipment, the historical mode selection information, and the historical execution situation of the tower crane operation tasks, use the A3C algorithm to construct an adaptive decision-making model for the tower crane equipment to select the auxiliary driving mode that best matches the tower crane equipment.

[0028] Step S103: Obtain the real-time status information of the tower crane equipment, and use the adaptive decision-making model to configure the target mode that matches the real-time status information.

[0029] In the embodiment of this application, the adaptive decision-making model is a model constructed based on the A3C (Asynchronous Advantage Actor - Critic) algorithm in deep reinforcement learning (DRL). This model can adaptively select the auxiliary driving mode most suitable for the current working conditions based on the historical data and real-time status information of the tower crane equipment. Its core lies in continuously learning the relationship between mode selection and task execution results in historical data, and automatically adjusting the strategy to achieve dynamic optimization of the operation mode.

[0030] In step S102, the A3C algorithm is a model-free deep reinforcement learning algorithm that combines policy gradient (Actor) and value function estimation (Critic). When constructing the adaptive decision-making model, historical state information of the tower crane equipment (such as hook position, boom angle, load weight, etc.), historical mode selection information (previously used assisted driving modes), and historical execution of tower crane operation tasks (whether the task was successfully completed, completion time, etc.) are used as training data. For example, through multi-threaded asynchronous training, multiple agents simultaneously learn and explore in different environment copies, continuously updating the parameters of the Actor network (used to generate action probability distributions) and the Critic network (used to evaluate state values), enabling the model to learn the strategy of selecting the optimal operation mode in different states.

[0031] Exemplarily, assume there is historical data of the past 1000 tower crane operations, including the state information of the tower crane during each operation (such as hook height, load weight, etc.), the selected assisted driving modes (joystick mode, voice driving mode, etc.), and the execution results of the operations (successfully completed the task, task timeout, etc.). These data are divided into a training set and a validation set and trained using the A3C algorithm. During the training process, the Actor network will output the probabilities of selecting different assisted driving modes based on the current state, and the Critic network will evaluate the value of the current state. By continuously adjusting the network parameters, the model can select the assisted driving mode that is most likely to lead to the successful completion of the task in different states.

[0032] Thus, by learning historical data, the model can better understand the applicability of various assisted driving modes in different states, assisting in the automation and intelligence of operation mode selection. With the continuous accumulation of historical data, the model can be continuously updated and optimized to adapt to different working conditions and task requirements.

[0033] Further optionally, in step S102, a variety of assisted driving modes are defined as an action space, and an assisted driving mode is selected from the action space at each time step; a reward function of the adaptive decision-making model is set based on the historical execution situation; historical state information, historical mode selection information, and historical execution situation of the tower crane equipment are obtained; multiple operation task environments are constructed according to the historical state information; the historical state information and historical mode selection information are input into the multiple operation task environments to parallelly train the Actor network for selecting the assisted driving mode; the historical state information, historical mode selection information, and historical execution situation are input into the multiple operation task environments to parallelly use the Critic network for evaluating the value of the Actor network; the network parameters of the Actor network and the Critic network are updated by using the reward function to obtain the adaptive decision-making model of the tower crane equipment. In the embodiment of the present application, the reward function at least includes: task completion reward, completion time reward, completion quality reward, and abnormal situation penalty of the tower crane equipment.

[0034] Specifically, in the scenario of tower crane assisted driving, multiple assisted driving modes such as the joystick mode, voice driving mode, fixed-distance driving mode, macro-control mode, and somatosensory control mode are defined as the action space. At each time step, the model needs to select an assisted driving mode from this action space to execute. For example, at a certain moment, the model has to decide whether to use the voice driving mode or the fixed-distance driving mode to control the tower crane. Furthermore, based on the historical execution of the tower crane operation tasks, a reward function is set. If the tower crane equipment successfully completes the lifting task, a positive reward is given as a task completion reward. For instance, accurately lifting the materials to the designated position can obtain a relatively high reward value; conversely, if the task fails, a penalty is given. To encourage quick task completion, if the actual task completion time is shorter than the expected time, an additional reward is given; if it exceeds the expected time, a penalty is imposed. When abnormal situations occur to the tower crane, such as hitting an obstacle or overloading, corresponding penalties are given to prevent such situations from happening again. Then, historical state information of the tower crane equipment is collected, such as the hook position, boom angle, load weight, etc.; historical mode selection information, that is, the assisted driving modes used in the past; and the historical execution of the tower crane operation tasks, such as whether the task is successful, completion time, completion quality, etc. These historical data are the basis for constructing and training the model. Subsequently, multiple operation task environments are constructed based on the historical state information. Each environment simulates different tower crane operation scenarios, such as different load weights, different operation spaces, different weather conditions, etc. This enables model training in multiple scenarios and improves the model's generalization ability. On the one hand, the historical state information and historical mode selection information are input into multiple operation task environments to parallel train the Actor network for selecting the assisted driving mode. The role of the Actor network is to output the probability distribution of selecting each assisted driving mode according to the current state information. Through continuous training in different operation task environments, the Actor network can learn which assisted driving mode should be selected in different states to obtain the maximum reward. On the other hand, the historical state information, historical mode selection information, and historical execution can be input into multiple operation task environments to parallel train the Critic network for evaluating the value of the Actor network. The main function of the Critic network is to evaluate the value of the action (assisted driving mode) selected by the Actor network in the current state. It will judge whether the current action is reasonable based on the reward value given by the reward function and provide feedback for the training of the Actor network. Finally, the reward function can be used to update the network parameters of the Actor network and the Critic network. During the training process, the model will continuously adjust the network parameters according to the feedback of the reward function, enabling the Actor network to more accurately select the assisted driving mode and the Critic network to more accurately evaluate the value of the action. After multiple iterative trainings, an adaptive decision-making model for the tower crane equipment is finally obtained.

[0035] It can be understood that, compared with the single-objective training in the traditional A3C algorithm, in the embodiment of the present application, in the tower crane assisted driving scenario, the reward function comprehensively considers multiple factors such as task completion, completion time, completion quality, and abnormal conditions, which is more in line with the actual needs of tower crane operations. Exemplarily, based on the specific design of the tower crane scenario, safety hard constraints (penalties for abnormal conditions) are added to preferentially avoid equipment damage or accidents, which meets the core requirements of industrial control. Or, the task quality rewards (precision, time) are subdivided to match the actual needs of "accurate lifting" and "efficient construction" in tower crane operations. At the same time, a mode switching penalty is also introduced to suppress policy oscillation and improve operation stability (the traditional algorithm may cause frequent mode switching due to excessive exploration). This can guide the model to pay attention to efficiency and quality while ensuring task completion, avoid abnormal conditions, and improve the practicability and safety of the model.

[0036] The traditional A3C algorithm has limitations in the processing of state space and action space. It usually only targets continuous actions or simple discrete actions (such as button operations in games), and the state space is mostly visual images or low-dimensional vectors, which are difficult to meet the complex requirements of the tower crane assisted driving scenario.

[0037] In the embodiment of the present application, the adaptive decision-making model in the tower crane assisted driving scenario further improves the adaptability to working conditions through engineering adaptation. Specifically, in the construction of the state space, multi-modal industrial data is fused, including equipment state data such as the hook position, load weight, and boom angle collected by sensors in real time, task characteristics such as the target position, precision requirements, and time limit in task attributes, and environmental parameters such as wind speed, obstacle distribution, and light intensity. And through feature engineering, key dimensions such as load weight and wind speed are normalized to form a high-dimensional state vector with physical meaning, enabling the model to accurately capture the complex working conditions of tower crane operation.

[0038] Further optionally, a self-supervised learning algorithm (such as SimCLRv2) can be introduced to enhance the sensor time-series data, automatically mine implicit information such as mechanical fatigue characteristics in the load fluctuation signal and risk associations after continuous emergency stops, and combine the Gaussian process regression model to predict the environmental change trend, so that the state vector is upgraded from a simple physical parameter splicing to an intelligent representation including time-series dependence and risk prediction. Through feature engineering processing such as load weight percentage normalization and wind speed Z-score standardization, the physical meaning of the data is further strengthened. At the same time, the digital twin technology is used to build a high-precision dynamic simulation environment to generate samples of extreme working conditions (such as ultra-limit wind speed and sudden load increase) missing in historical data, and supplement the coverage of dangerous scenarios in training data, enabling the model to accurately capture complex working condition characteristics such as the coupling effect of boom angle and wind speed and the dynamic effect of load inertia on hook swing.

[0039] In the design of the action space, it breaks through the direct processing of underlying control actions such as motor speed and joint angle in traditional algorithms, abstracting them into discrete assisted driving mode selections, with each mode encapsulating mature upper-layer control strategies. For example, the voice mode integrates a semantic parsing model based on Transformer to achieve multi-instruction context understanding, and the fixed-distance mode embeds a PID parameter self-tuning algorithm combined with visual feedback to improve positioning accuracy. On this basis, the meta-learning (MAML) technology is introduced to initialize the policy parameters for different task types (heavy-load lifting, precision positioning), enabling the model to quickly adapt to the dynamic parameter changes of new types of tower cranes through a small number of samples. At the same time, a three-level redundant decision-making mechanism of the main policy (A3C generation mode), backup policy (heuristic rules), and expert rules (safety bottom line) is constructed, and the Dempster-Shafer evidence theory is used to fuse the decision-making results. When critical working conditions such as wind speed > 20m / s are detected, the coordinated control of the anti-wind mode and the safety fallback strategy is automatically triggered. In this way, while retaining the adaptive ability of reinforcement learning, hard-coded safety rules are used to prevent the model from outputting dangerous decisions. This kind of abstraction processing and policy-level encapsulation reduces the action space dimension from dozens of underlying control parameters in traditional algorithms to 5 mode selections. While reducing the number of model parameters, it also improves the decision-making reliability, significantly reduces the difficulty of engineering deployment, and enhances the adaptability to actual scenarios.

[0040] Exemplarily, in the tower crane operation at a bridge construction site, the improved technology for constructing the state space significantly enhances the model's understanding ability of complex working conditions through multi-dimensional data fusion and intelligent processing. When the tower crane continuously lifts a 50-ton box girder, the load fluctuation signal collected by the sensor is processed by SimCLRv2 self-supervised learning, and the periodic load abnormal fluctuation characteristics caused by slight wear of the steel wire rope are automatically identified. This characteristic is easily overlooked in traditional manual feature engineering, and the adaptive decision-making model discovers its 89% correlation with the steel wire rope breakage accident within the next 2 hours through contrastive learning, thus triggering an equipment maintenance warning in advance.

[0041] Meanwhile, the adaptive decision-making model can also be processed through Gaussian process regression. Based on historical wind speed data and real-time meteorological radar signals, it predicts that strong winds of 18 m / s will be encountered in the next 30 minutes. The system automatically incorporates parameters such as "wind speed change trend", "current angle of the boom", and "load moment of inertia" into the state vector dynamically, enabling the model to be pre-adjusted to the wind-resistant mode 10 minutes before the strong wind arrives. Further optionally, a compound working condition of ultra-limit wind speed with missing historical data and full load can also be generated through digital twin to simulate the following scenarios: simulating the critical instability state of the boom when the wind speed is 25 m / s. The adaptive decision-making model learns through this virtual data the strategy that "when the product of the wind speed and the boom angle exceeds the safety threshold, even if the load limit is not reached, staged unloading is required", filling the decision-making gap for dangerous working conditions that are difficult to verify in real scenarios.

[0042] When the operator issues a voice command "Lift the precast slab to a height of 20 meters, move to the coordinates (15, 8, 20) and then fine-tune and place it", the voice mode parsing module integrated with Transformer not only recognizes a single command, but also captures the precise positioning intention behind this "fine-tuning" command, and automatically combines the fixed-distance mode and the macro-control mode. In this way, in the fixed-distance mode, the moving path can be controlled with an accuracy of 1 cm through a PID algorithm with embedded visual feedback. In the macro mode, a vibration compensation algorithm is enabled during the last 50 cm of travel, reducing the placement error of the component from ±5 cm in the traditional solution to ±1.2 cm.

[0043] Facing a newly arrived tower crane, assuming that the length of the boom is 15% longer than the old model, the adaptive decision-making model can also quickly adjust the path planning parameters of the fixed-distance mode based on the boom angle and speed data collected during 3 trial lifts of the new equipment through the MAML algorithm, shortening the completion time of the new equipment's first task from 40 minutes in traditional transfer learning to 12 minutes, and maintaining the same positioning accuracy.

[0044] In the practical application of the three-level redundant decision-making mechanism, when the tower crane performs heavy-load lifting at a wind speed of 16 m / s, the main strategy tends to select the joystick mode because there are few samples in this wind speed range in historical data. However, its risk assessment score (0.65) is lower than the heuristic rule score of the backup strategy (0.82, rule: give priority to the automatic mode when the wind speed > 15 m / s). Combining the safety bottom line of the expert rule (manual high-speed slewing is prohibited when the wind speed > 12 m / s), the adaptive decision-making model finally triggers the coordinated control of the wind-resistant mode and the fixed-distance mode, restricting the slewing speed of the boom to 0.3° / s.

[0045] Taking the hoisting of precast beams with different weights by tower cranes (such as 70-ton T-beams and 30-ton hollow slabs) as an example, each assisted driving mode encapsulates an independent control strategy. The fixed-distance mode is equipped with a PID parameter library based on the load. For example, the PID parameters corresponding to a 70-ton load are: P = 0.8, I = 0.3, D = 0.5. The wind resistance mode dynamically adjusts the elevation angle of the boom according to the wind speed, gravitational acceleration, and boom length through a formula. Among them, the wind speed is collected in real time by a sensor, the gravitational acceleration is about 9.8 m / s² as a constant, and the boom length is determined according to the tower crane model and working conditions. By adjusting the elevation angle, the force distribution of the tower crane is changed. Increasing the elevation angle in strong winds can reduce the lateral wind pressure, avoid overturning, and ensure the stability of the tower crane.

[0046] Meta-learning can play a role in accelerating the strategy in this architecture. First, the Meta-Learner extracts the meta-parameters of each mode from historical hoisting tasks. Taking the task of "for every 10-ton increase in load, the PID-P parameter of the fixed-distance mode increases by an average of 0.15" as an example, and encodes it as meta-knowledge. When a new type of tower crane (the boom length L increases from 45 meters to 55 meters) first hoists an 80-ton box girder, the Meta-Learner, based on the encapsulated wind resistance mode strategy framework, uses the meta-knowledge of the "mapping relationship between L and θ" in historical tasks to directly generate initial control parameters (referring to the aforementioned formula, θ can be obtained to be approximately 19.7°), reducing the number of trial-and-error times by 70% compared to traditional random initialization.

[0047] In more complex composite working conditions (such as the "strong wind + variable load" scenario), the strategy units encapsulated by the mode become the basic operation units of meta-learning. When the wind speed suddenly increases from 12 m / s to 18 m / s and the load dynamically adjusts from 60 tons to 40 tons, the meta-learning algorithm (such as MAML) no longer optimizes the underlying motor control parameters, but performs joint parameter fine-tuning for the encapsulated wind resistance mode and fixed-distance mode, that is: through 10 virtual environment trainings (taking less than 2 seconds), quickly determine that the boom angle compensation coefficient of the wind resistance mode is adjusted from 0.9 to 1.2, and the path planning speed threshold of the fixed-distance mode is increased from 0.5 m / s to 0.7 m / s, increasing the first decision accuracy rate of the model in the new working condition from 30% to 85%. This combination method essentially transforms the domain knowledge encapsulated by the mode into a transferable strategy prior of meta-learning, retaining the professionalism of each mode (such as the physical model constraints of the wind resistance mode), and realizing knowledge transfer between modes through meta-learning (such as the correlation law of mode parameters under different loads). Finally, in a certain construction project, the strategy adaptation time of the new equipment is shortened from 48 hours to 1.5 hours, and the mode switching failure rate under complex working conditions is reduced by 60%.

[0048] In this way, the policy parameters encapsulated in each mode (such as PID coefficients, semantic parsing model weights) constitute subtasks of meta-learning. The meta-learner optimizes the initialization parameters of these subtasks through gradient descent, enabling it to quickly converge with only a small number of samples in new tasks. For example, in the semantic parsing model of the voice mode, meta-learning discovers that there are commonalities in the word vector mappings of instruction features related to hoisting accuracy requirements (such as keywords like precise and fine-tuning) in different projects. Therefore, when deploying at a new construction site, the parsing model can be adapted with fewer new instruction samples, reducing the need for labeled data by 80% compared to traditional fine-tuning methods.

[0049] The above example avoids the complexity of directly operating on the underlying control actions. By endowing the model with policy-level transfer capabilities through meta-learning, the tower crane assisted driving system demonstrates industrial-level fast adaptation capabilities in the face of equipment iteration and working condition changes.

[0050] In the embodiments of this application, the adaptive decision-making model supports an online fine-tuning mechanism. When new working condition data such as data of new model tower cranes and data of operating in extreme weather accumulates to a certain scale, the model parameters are updated through incremental training instead of retraining the entire network, significantly improving the model's adaptability to new scenarios and training efficiency.

[0051] Aiming at the problem of gradient oscillation caused by environmental differences in traditional A3C multi-threaded asynchronous training, the adaptive decision-making model has achieved a double improvement in training stability and efficiency through an engineering acceleration strategy. Exemplarily, in the embodiments of this application, the adaptive decision-making model clusters the environmental copies according to task types, divides the complex tower crane operations into typical scenarios such as "heavy load hoisting", "high-precision positioning", and "operating in strong wind environment", and each worker thread focuses on policy training under specific working conditions, reducing the interference caused by the mixing of data in different scenarios and improving the sample utilization efficiency.

[0052] Furthermore, in the embodiments of this application, a priority experience replay mechanism is also introduced, which assigns higher weights to the trajectories of high-reward events (such as precise operations that successfully avoid obstacles) and high-risk events (such as safety control under critical overloading conditions). By preferentially learning these key experiences, the convergence of the model to safety policies and efficient policies is accelerated, avoiding the problem of dilution of important experiences caused by random sampling in traditional algorithms, and making the training process more targeted to optimize key decision-making logics.

[0053] After the adaptive decision-making model is constructed, in step S103, the real-time state information of the tower crane equipment is obtained, including the current hook position, boom angle, load weight, and environmental parameters (such as wind speed, light, etc.). These real-time state information is input into the trained adaptive decision-making model, and the model outputs the auxiliary driving mode most suitable for the current state, that is, the target mode.

[0054] Exemplarily, at a certain moment, the real-time status information of the tower crane is obtained: the hook height is 10 meters, the load weight is 5 tons, and the wind speed is 3 m / s. These information are input into the adaptive decision-making model, and the model judges according to the internal strategy that the currently most suitable auxiliary driving mode is the fixed-distance driving mode. Because in this state, the fixed-distance driving mode can more accurately control the movement of the hook to complete the current operation task. Thus, the target mode is dynamically selected according to the real-time status information to ensure that the tower crane can adopt the most suitable operation mode at any time, quickly adapt to different working conditions changes and adjust the operation mode in time, improve the operation efficiency and safety, and ensure the stable operation of the tower crane.

[0055] As an optional embodiment, in step S103, configuring the target mode matching the real-time status information by using the adaptive decision-making model includes: obtaining the image data around the tower crane equipment, and performing obstacle positioning based on the image data to obtain the obstacle positioning data of the tower crane equipment; parsing the real-time status information to obtain the real-time equipment component status, real-time weather condition, and real-time wind speed condition of the tower crane equipment; inputting the real-time equipment component status, real-time weather condition, real-time wind speed condition, and obstacle positioning data into the adaptive decision-making model, and using the Actor network to predict the driving mode to obtain the prediction probability of the tower crane equipment in each auxiliary driving mode; randomly selecting a candidate mode by roulette based on the prediction probability; monitoring the safety risk of the tower crane equipment in the candidate mode; if the safety risk is greater than the set threshold, setting the target mode to the auxiliary driving mode matching the safety risk item, or switching the target mode to the auxiliary driving mode with a safety level higher than the current target mode.

[0056] In principle, first, the image data around the equipment is collected by using a camera, and the image is processed through computer vision algorithms (such as YOLO, Mask R-CNN) to realize the recognition and positioning of obstacles and obtain the obstacle positioning data. At the same time, the real-time status information of the tower crane is parsed, including the equipment component status (such as the hook position, boom angle), weather condition, wind speed condition, etc., to provide comprehensive data support for subsequent decision-making. Then, the above data are input into the Actor network of the adaptive decision-making model, and the network predicts each auxiliary driving mode to obtain the probability of selecting each auxiliary driving mode. Based on the prediction probability output by the Actor network, a candidate mode is randomly selected by roulette. After determining the candidate mode, the safety risk of the tower crane operation is monitored in real time. If the risk exceeds the threshold, the target mode is set to the mode matching the safety risk according to the risk type, or switched to a higher safety level mode to ensure the safe operation of the tower crane.

[0057] Exemplarily, assume that at a construction site, a tower crane is performing a hoisting operation. In this case, it can be photographed by a camera that there are some building materials and construction equipment around the tower crane. After obstacle positioning, it is determined that there is a pile of steel bars 3 meters away from the tower crane's boom, which may obstruct the slewing operation of the tower crane. Assume that the real-time state of the equipment components is that a 5-ton heavy object is hung on the hook, the boom angle is 45 degrees, and it is in the process of slewing. Assume that the real-time weather condition is that it is about to rain. Assume that the real-time wind speed condition is that the current wind speed is 12 m / s and there is a tendency to gradually increase. Based on these assumptions, after inputting this information into the Actor network of the adaptive decision-making model, the predicted probabilities of each assisted driving mode are obtained as follows: the probability of the joystick mode is 0.3, the probability of the voice driving mode is 0.2, the probability of the fixed-distance driving mode is 0.25, the probability of the macro-control mode is 0.15, and the probability of the somatosensory control mode is 0.1. Through roulette wheel random selection, the candidate mode is determined to be the fixed-distance driving mode. During the operation of the tower crane in the fixed-distance driving mode, it is monitored that as the wind speed continuously increases to 15 m / s and the boom approaches the obstacle, there is a high risk of collision and overturning, and the safety risk exceeds the set threshold. In this case, according to the safety risk situation, the target mode is switched to a mode combining the wind resistance mode and the macro-control mode: the wind resistance mode is used to cope with strong winds and enhance the stability of the tower crane by adjusting the boom angle and operating parameters; the macro-control mode is used to perform precise fine operations when approaching the obstacle to avoid collisions and ensure the safe completion of the hoisting operation.

[0058] In this way, the tower crane assisted driving system dynamically adjusts the target mode by real-time monitoring of safety risks, reduces the probabilities of accidents such as collisions and overturnings, and ensures the safety of personnel and equipment. By fusing multi-source data to predict the driving mode and flexibly selecting the mode according to the environment, equipment state, and weather, the adaptability to different scenarios is improved. The combination of roulette wheel random selection and risk monitoring not only ensures reasonable decision-making but also avoids conservative and single mode selection, improving the decision-making and operation efficiency.

[0059] Step S104: Receive the operation instruction issued by the user to the tower crane equipment, and construct a tower crane task model of the tower crane equipment in the target mode based on the operation instruction.

[0060] The tower crane task model is a model constructed for the tower crane equipment to execute tasks based on user operation instructions and target modes. This model can decompose and plan the entire operation process of the tower crane to ensure that the tower crane can accurately and safely complete tasks according to the predetermined steps and requirements. According to the user's operation instructions and the current target mode, the overall operation of the tower crane is planned, and the tasks to be completed in each stage are clarified, so that the operation of the tower crane has a clear process and sequence, avoiding chaos and errors. This model coordinates the work between various components and systems of the tower crane to ensure that in different task stages, each component can work together as required to achieve precise position control, speed control, force control, etc., in order to complete specific operation tasks such as lifting goods and adjusting the position of the boom. Through the reasonable planning of tasks and the strict control of each task node, this model can take various safety factors into account in advance, set corresponding safety inspections and protection mechanisms, reduce safety risks during the operation process, and ensure the safety of the tower crane equipment and personnel. This model can be used to optimize the operation process, reduce unnecessary actions and time waste, improve the working efficiency of the tower crane, enable the tower crane to complete tasks in the shortest time, and improve the construction progress.

[0061] Among them, multiple task nodes are sequentially set in the tower crane task model according to the operation time sequence from first to last, and each task node corresponds to an operation task to be executed. These task nodes are interrelated and executed sequentially, jointly constituting the complete process for the tower crane to complete an operation task. Each task node has clear task objectives and operation requirements. Through the precise control and coordination of these task nodes, it can be ensured that the tower crane safely and efficiently completes various operation tasks.

[0062] Exemplarily, in the tower crane task model, the data information that each task node can contain is as follows: In task node 1, in the preparation stage, check the status of the tower crane equipment components (hook, boom, wire rope, etc.), confirm whether the load weight exceeds the rated load, and check whether there are obstacles in the surrounding environment; at the same time, according to the target mode and operation instructions, initialize the tower crane control system and set parameters such as the initial angle of the boom and the initial position of the hook. In task node 2, in the lifting stage, according to the load weight and the target position, calculate and control the appropriate lifting speed and force, and slowly lift the hook to lift the load off the ground. During the lifting process, monitor the status of the load and the stability of the tower crane in real time to ensure a smooth lifting process and avoid load sway or tower crane tilt. In task node 3, in the horizontal movement stage, after the load leaves the ground and reaches a certain height, control the boom to perform slewing and luffing operations to horizontally move the load above the target position. During this process, it is necessary to accurately control the movement speed and angle of the boom according to factors such as wind speed, boom length, and load weight, and at the same time maintain the balance and stability of the load to prevent the load from swinging too much due to inertia or wind force. In task node 4, in the lowering stage, when the load reaches above the target position, control the tower crane to slowly lower the hook height and accurately place the load at the specified position. During the lowering process, closely monitor the distance and alignment between the load and the target position, and timely adjust the lowering speed and hook position to ensure that the load can be placed smoothly and accurately. In task node 5, in the final stage, after the load is placed, raise the hook and return it to its original position, turn off the power system of the tower crane, check and maintain the tower crane equipment, and clean up the site. At the same time, record the relevant data of this operation, such as operation time, load weight, boom movement parameters, etc., for subsequent data analysis and equipment management.

[0063] Exemplarily, the operation instruction is at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a somatosensory control instruction. Specifically, the linkage console operation instruction includes the action information of each mechanism of the tower crane (such as hoisting, luffing, slewing, etc.), such as the rotation speed and direction of the motor, as well as the action amplitude and position data of each mechanism. It may also include the timestamp of the operation to record the moment when the operation occurs for sequential control and time-related analysis. The voice instruction includes the audio data of the voice, which is stored and transmitted in the form of digital signals after being encoded. In addition, it includes the parsed semantic information, such as specific operation commands (such as rising, falling, turning left, turning right, etc.), the target object of the operation (such as the lifted goods, the moving position, etc.), and possible modifiers (such as slowly, quickly, etc.). It may also include auxiliary information such as the source and time of the voice instruction. The fixed-distance driving instruction mainly includes distance-related data, such as the horizontal distance and vertical distance that the tower crane needs to move, and the coordinate information of the target position. It may also include speed information, that is, the speed at which the tower crane is expected to move to the specified distance, and acceleration limit data to ensure the smoothness and safety of the tower crane during movement. There will also be information such as the execution time of the instruction. The macro-control instruction includes data related to fine operations, such as the precise control parameters of the motor, such as the adjustment values of current and voltage, to achieve small-amplitude actions. It also includes position feedback information for precisely monitoring the current position and state of the tower crane for real-time adjustment. In addition, it may include data such as the accuracy requirements of the operation, the small offset of the target position, and time parameters related to the macro-operation, such as the duration and interval time of the operation. The somatosensory control instruction includes the posture data of human movements, such as the shape, angle, and position changes of gestures. These data are collected by sensors (such as cameras, somatosensory bracelets, etc.) and converted into digital signals. It also includes the time series information of the actions to record the sequence and duration of the gesture actions for learning the dynamic changes of the gestures. Additionally, it may include identification information related to the somatosensory device, as well as the start and end times of the operation.

[0064] As an optional embodiment, in step S104, from the candidate algorithm model library, select the target parsing model that matches the operation instruction; use the target parsing model to process the operation instruction to obtain the operation instruction feature vector; the operation instruction feature vector includes the instruction sequence between each operation task, the task type to which the operation task belongs, and the logical relationship; based on the operation instruction feature vector and the pre-set association relationship between the task type and the task priority, determine the operation tasks to be executed and the arrangement method of the operation tasks; according to the operation tasks to be executed and the arrangement method of the operation tasks, combine them through a graph neural network to obtain the tower crane task model.

[0065] Specifically, the candidate algorithm model library stores a variety of parsing models for different operation instructions. Based on the input operation instruction type, a matching target parsing model is selected from the library. For example, for voice instructions, a model combining BERT, Transformer, and self-attention is selected; for motion control instructions, a model combining DCNN and LSTM is selected. The target parsing model then processes the operation instruction, converting it into an operation instruction feature vector. This feature vector contains important information such as the instruction sequence, task type, and logical relationships between each task. For example, for a multi-step voice instruction such as "First, lift the hook, then move it to the specified position, and finally, lower the hook," the feature vector records the order of these steps and the task type (e.g., lift, move, lower, etc.) for each step. Based on the operation instruction feature vector and the pre-set relationship between task type and task priority, the task to be executed and its arrangement are determined. Different task types may have different priorities; for example, safety-related tasks typically have higher priority. Based on these priorities and instruction sequence, the execution order and method of each task are determined. Finally, based on the tasks to be performed and their arrangement, a graph neural network is used to construct a crane task model. Graph neural networks can effectively handle the complex relationships between tasks, representing each task and its relationships as a graph structure, thereby constructing a complete crane task model. This model provides clear guidance for crane operation, enabling the crane to perform tasks according to predetermined steps and requirements.

[0066] Further optionally, parsing models may be pre-built for each of these operation instructions and stored in a candidate algorithm model library.

[0067] For example, the linkage platform operation instructions, as structured control signals, need to be parsed into the coordinated control logic of the tower crane actuators, focusing on the timing constraints and safety thresholds of multi-mechanism actions. To this end, the state machine-graph neural network maps the instructions to mechanism state transitions. Using the graph neural network to model the spatial dependencies between mechanisms, it replaces the traditional rule engine and automatically learns linkage rules in complex scenarios. Optionally, a decision tree enhanced by an attention mechanism is used to dynamically weight the instruction timing data, generate a priority decision tree, and detect abnormal operation combinations in real time, improving operational safety and error detection rate.

[0068] Exemplarily, voice command parsing needs to extract operation intent, target object, and constraints from voice signals to achieve end-to-end processing of speech recognition and semantic parsing. Multimodal pre-training models such as UniSpeech and SpeechT5 are pre-trained by combining voice waveforms and text semantics, directly converting voice commands into structured operation parameters, supporting the parsing of complex semantics in long sentences, and reducing intermediate error transmission. The speech-semantic alignment model enhanced by contrastive learning aligns voice features and semantic vectors through contrastive learning, solves the ambiguity of homophones, and the recognition accuracy is increased by more than 30% in the noisy environment of the construction site, and it can also be compatible with dialect parsing.

[0069] Exemplarily, fixed-distance driving command parsing aims to convert movement commands into path planning and trajectory optimization, while satisfying constraints such as speed and acceleration. Deep model predictive control combines deep learning and MPC, uses neural networks to model the non-linear dynamic characteristics of tower cranes, adds safety margin constraints, dynamically adapts to environmental changes, and reduces trajectory tracking errors. The reinforcement learning-model predictive control fusion pre-trains strategies through offline reinforcement learning and generates short-term trajectories online using MPC, balancing global optimality and real-time performance to achieve fast path adjustment in high-speed scenarios.

[0070] Exemplarily, macro manipulation command parsing focuses on converting high-precision operation commands into fine control of the motor servo system, processing high-frequency position feedback and dynamic disturbances. Neural differential equations model motor control as a continuous-time dynamic system, learn differential equation parameters to achieve continuous control of current and voltage, and reach microsecond-level control accuracy to meet the requirements of high-precision docking. Hierarchical reinforcement learning decomposes operations into task layers and control layers, optimizes strategies hierarchically, reduces the state space dimension, and improves the learning efficiency and control stability in complex scenarios such as multi-axis collaborative fine-tuning.

[0071] Exemplarily, somatosensory manipulation command parsing needs to recognize operation intent from the spatio-temporal sequence of human actions. The spatio-temporal graph neural network models gesture skeleton points as a graph structure, extracts spatial joint associations and time series dynamics through spatio-temporal graph convolution, supports the recognition of complex continuous actions, and is robust to occlusion and perspective changes. The video Transformer uses the self-attention mechanism to capture the action dependencies between somatosensory video frames, learns action patterns end-to-end, and achieves low-latency response.

[0072] Furthermore, in terms of the construction and adaptation of the candidate algorithm model library, it stores and annotates applicable scenarios by instruction type classification, supports dynamic loading and online updating, and converges multi-site data through federated learning to optimize the generalization ability. The model matching strategy automatically selects the optimal parsing model based on real-time status information such as weather, light, and noise. For example, it switches to the ST-GNN model based on a depth camera for somatosensory control under strong light, and selects the noise-resistant enhanced CL-SAM model for voice parsing in a high-noise environment, thus realizing the full-link intelligence from instruction input to precise execution, and significantly improving the operation safety and efficiency of tower cranes under complex working conditions.

[0073] Further optionally, after receiving the operation instruction issued by the user to the tower crane equipment in step S104, the operation instruction can also be parsed to obtain the operation task requirements in the operation instruction; it is judged whether the current target mode of the tower crane equipment meets the operation task requirements. If the target mode does not meet the operation task requirements, the target mode is reselected based on the operation instruction; the operation instruction, the real-time operation information of the tower crane equipment, and the reselected target mode are input into the fuzzy control model to generate the equipment control parameters during the target mode switching process, so as to smoothly switch the tower crane equipment from the current target mode to the reselected target mode. The equipment control parameters at least include: the motor speed adjustment value, the braking force of the braking device, and the speed of the luffing mechanism.

[0074] In the above steps, after receiving the user operation instruction, first extract the core operation task requirements in the instruction through technologies such as natural language processing and signal parsing, such as lifting weight, target position, accuracy requirements, time limit, etc. At the same time, combine the current target mode of the tower crane equipment (such as the joystick mode, the fixed-distance driving mode, etc.) and its inherent capabilities (such as the fixed-distance mode is good at precise movement, and the joystick mode has high flexibility but low accuracy), and compare and analyze whether the target mode can meet the operation task requirements. For example, if the operation task requires placing a heavy object at a specific position with millimeter-level accuracy, and the current target mode is the joystick mode, since its accuracy is difficult to meet the requirements, it is determined that the current mode does not meet the task requirements.

[0075] When it is determined that the current target mode does not meet the requirements of the operation task, based on the key information in the operation instruction, referring to historical data and preset mode selection rules, a suitable target mode is reselected. For example, if the current task is high-precision lifting, the system may switch the target mode from the joystick mode to the fixed-distance driving mode or the micro-distance control mode to better match the high-precision operation requirements. Furthermore, the operation instruction, the real-time operation information of the tower crane equipment (such as the current hook position, boom angle, load weight, etc.), and the reselected target mode are input into the fuzzy control model. The fuzzy control model performs fuzzy processing on the input information through preset fuzzy rules, converting the accurate input data into fuzzy sets (such as fuzzy concepts like "fast speed" and "large braking force"). Then, according to the fuzzy inference rules, logical reasoning is performed on the fuzzy sets to obtain a fuzzy output. Finally, through defuzzification operations, the fuzzy output is converted into accurate equipment control parameters, such as the motor speed adjustment value, the braking force of the braking device, and the speed of the luffing mechanism. These parameters are used to control the tower crane equipment during the mode switching process to achieve a smooth and safe transition, avoiding mechanical shocks or operation out-of-control caused by mode switching.

[0076] Exemplarily, assume that at a certain construction site, the tower crane is currently in the joystick mode, and the operator issues an operation instruction "Precisely lift a 5-ton heavy object to a specified platform 20 meters away and 15 meters high from the current position". After parsing the instruction, it is clear that the operation task requirement is to lift a 5-ton heavy object to a specific position with high precision requirements. Evaluating the current joystick mode, it is found that it is difficult to meet the high-precision lifting requirements, and it is determined that the current target mode does not meet the operation task requirements. According to the operation task requirements, the system reselects the fixed-distance driving mode as the target mode because this mode can better achieve precise distance control and positioning. Furthermore, the operation instruction, the current real-time operation information of the tower crane (such as the current hook position, the current boom angle, and the load weight of 5 tons), and the reselected fixed-distance driving mode are input into the fuzzy control model. The model processes according to the preset fuzzy rules. For example, considering that the load is heavy and precise movement is required, the equipment control parameters generated by the fuzzy control model may be: the motor speed adjustment value is set to a lower speed to ensure a smooth lifting process; the braking force of the braking device is appropriately increased to prevent sliding during the lifting process; the speed of the luffing mechanism is also set to a lower value to ensure that the position of the boom can be precisely adjusted. Through these control parameters, the tower crane smoothly switches from the joystick mode to the fixed-distance driving mode and successfully completes the lifting task.

[0077] Through dynamic evaluation and target mode reselection, the tower crane automatically adapts its operating mode to operational needs, improving its adaptability to diverse scenarios and avoiding task obstructions and efficiency losses. Parameters generated by the fuzzy control model ensure smooth mode switching, reducing mechanical shock, enhancing operational stability, and extending equipment life. Precise mode selection and stable switching reduce safety risks and prevent accidents such as overloading and collisions. Automatically matching the optimal mode and switching quickly and smoothly reduce manual intervention, improving tower crane operating efficiency, accelerating construction progress, and reducing costs.

[0078] As an optional embodiment, it is assumed that the real-time operation information includes at least: the boom position of the tower crane equipment, the hook position, the equipment motion state information, the load information, and the environmental information.

[0079] Based on the above assumptions, in the above steps, the operating instructions, the real-time operation information of the tower crane equipment, and the re-selected target mode are input into the fuzzy control model to generate the equipment control parameters in the target mode switching process, so as to smoothly switch the tower crane equipment from the current target mode to the re-selected target mode, including: taking the operation target, operation type, real-time operation information, and re-selected target mode in the operating instructions as input variables, identifying the value range and actual physical meaning of each input variable, and configuring a corresponding first fuzzy set and membership function for each input variable based on the identification result; calculating the membership of each input variable in the corresponding first fuzzy set through the membership function; performing fuzzy reasoning on the target mode switching process of the tower crane equipment based on the membership and pre-set fuzzy rules to obtain a second fuzzy set for constructing the equipment control parameters; performing defuzzification processing based on the second fuzzy set to obtain the equipment control parameters.

[0080] Specifically, in the above steps, the operation objectives (such as lifting to a specific position, placement accuracy requirements, etc.), operation types (such as ordinary lifting, precision installation, etc.) in the operation instructions, as well as the real-time operation information (boom position, hook position, equipment motion state information, load information, environmental information) and the reselected target mode are used as input variables. First, identify the value range and actual physical meaning of each input variable. Then, configure the corresponding first fuzzy set and membership function (such as triangular membership function, trapezoidal membership function, etc.) for each input variable according to these identification results. The membership function is used to describe the degree to which an input variable belongs to a certain fuzzy set. Next, calculate the membership degree of each input variable in its corresponding first fuzzy set through the membership function. For example, if the boom angle is 60 degrees, calculate its membership degree in the fuzzy set of "small boom angle" according to the membership function, and so on, calculate the membership degrees of all input variables in their respective fuzzy sets. Based on the calculated membership degrees and the preset fuzzy rules (such as "if the boom angle is large and the load is heavy, then the adjustment value of the motor speed should be small" and other rules), perform fuzzy reasoning on the target mode switching process of the tower crane equipment. Through these rules, perform logical operations on the membership degrees of the input variables to obtain the second fuzzy set (such as "small adjustment value of the motor speed", "large braking force of the braking device", etc.) used to construct the equipment control parameters. Finally, perform defuzzification processing on the obtained second fuzzy set to convert the fuzzy result into an accurate numerical value, that is, obtain the equipment control parameters (such as the specific numerical value of the motor speed adjustment value, the specific magnitude of the braking force of the braking device, the specific value of the luffing mechanism speed, etc.).

[0081] Exemplarily, assume that the tower crane is currently in the linkage console mode, and the operator issues an operation instruction "Precisely hoist an 8-ton heavy object to a designated platform 30 meters horizontally and 20 meters vertically from the current position". After system evaluation, the fixed-distance driving mode is re-selected as the target mode. Assume that the operation target is to hoist to a position 30 meters horizontally and 20 meters vertically, with high precision requirements. Assume that the operation type is precision hoisting. Assume that the real-time operation information is as follows: the boom angle is 45 degrees (value range 0 to 180 degrees), and fuzzy sets "small boom angle", "medium boom angle", and "large boom angle" are configured; the hook position is 10 meters horizontally and 8 meters vertically from the target position, and corresponding fuzzy sets are configured; the equipment motion state information such as the hook lifting speed is 0.5 m / s (value range 0 to 2 m / s), and fuzzy sets such as "slow speed", "medium speed", and "fast speed" are configured; the load information is a load weight of 8 tons (exceeding the medium load in general cases), and fuzzy sets "light load", "medium load", and "heavy load" are configured; the environmental information such as the wind force is level 4 (value range 0 to 12 levels), and fuzzy sets "small wind force", "medium wind force", and "large wind force" are configured. Assume that the target mode is the fixed-distance driving mode. Based on the above assumptions, in this example, a corresponding membership function is configured for each fuzzy set. For example, for the "medium boom angle" fuzzy set of the boom angle, a triangular membership function is used. Through calculation using the membership function, it is obtained that, for example, the membership degree of the boom angle of 45 degrees in the "medium boom angle" fuzzy set is 0.6; the membership degree of the hook position in the corresponding fuzzy set; the membership degree of the load weight of 8 tons in the "heavy load" fuzzy set, etc. According to the pre-set fuzzy rules, such as "If the load is heavy and the boom angle is medium, and at the same time the wind force is medium, then the adjustment value of the motor speed should be small and the braking force of the braking device should be large", reasoning is carried out in combination with the calculated membership degrees to obtain a second fuzzy set for constructing equipment control parameters, such as "small adjustment value of the motor speed", "large braking force of the braking device", etc. The center-of-gravity method is used to defuzzify the second fuzzy set to obtain equipment control parameters such as the motor speed adjustment value of 0.2 m / s, the braking force of the braking device as a specific value, and the luffing mechanism speed as a specific value, thereby realizing a smooth switch from the linkage console mode to the fixed-distance driving mode.

[0082] Thus, the fuzzy control model comprehensively processes various input variables such as the operation target, type, real-time information, and target mode, generates equipment control parameters, ensures a smooth transition of the tower crane during mode switching, reduces mechanical impact and operation fluctuations, and improves stability and reliability. It can flexibly adjust parameters according to different working conditions and operation requirements, enhance the adaptability of the tower crane to complex environments, and improve operation safety and efficiency. In precision operations, it can accurately adjust parameters to meet high-precision requirements. In addition, the model simulates expert decision-making based on pre-set fuzzy rules, provides intelligent support for tower crane operation, reduces manual intervention and decision-making errors.

[0083] Further optionally, in the above steps, defuzzification is performed based on the second fuzzy set to obtain device control parameters, including: using the centroid method, according to the membership degrees and value ranges of the fuzzy elements in the second fuzzy set, determining the centroid position of the second fuzzy set, and taking the value corresponding to the centroid position as the first device control parameter. Using the maximum membership degree method, according to the membership degrees and value ranges of the fuzzy elements in the second fuzzy set, selecting the fuzzy element with the maximum membership degree, and taking the midpoint of the value range of the selected fuzzy element as the second device control parameter. Based on the data value characteristics of the device control parameters, the final output device control parameter is selected from the first device control parameter and the first device control parameter. The following takes the combined use of the two methods as an example to illustrate the specific implementation process.

[0084] The centroid method is a defuzzification method that determines the result by calculating the weighted average of all elements in the fuzzy set (the weight is the membership degree of the element). This method comprehensively considers all element information in the set, can more comprehensively reflect the overall characteristics, is more applicable in scenarios with high control accuracy requirements, can accurately determine device control parameters, and reduce errors. It is applicable to three types of scenarios: First, when multiple factors need to be comprehensively considered, such as in the complex operation of a tower crane, multiple factors such as the hook position, speed, and load act together on the lifting speed adjustment, and the fuzzy state contributions of various factors can be fully utilized. Second, in scenarios with strict control accuracy requirements, like the installation of large building components, it can ensure precise and safe operation. For example, in large building projects, the accuracy requirements for the installation position of components are extremely high, and using the centroid method can better meet the control accuracy requirements and ensure the accuracy and safety of tower crane operations. Third, in the face of a uniform or complex distribution of the membership function, when there is no obvious dominant element in the fuzzy set, a reasonable result can be obtained based on the actual distribution, such as the complex distribution of speed adjustment values in different operation stages of a tower crane. For example, in different operation stages of a tower crane, the membership degrees of each fuzzy state in the fuzzy set of the lifting speed adjustment value may show a complex distribution due to various factors, and at this time, the centroid method can calculate a suitable adjustment value according to the actual distribution.

[0085] In an example of a calculation process of the centroid method, assume that there are three fuzzy elements regarding the motor speed adjustment value in the second fuzzy set, namely "increase", "remain unchanged", and "decrease", and their membership degrees and value ranges are as follows: Increase: The membership degree is 0.3, and the value range is [5%, 10%], and the representative value in this range can be set as 7.5% (usually taking the midpoint). Remain unchanged: The membership degree is 0.4, and the value range is [-1%, 1%], and the representative value is 0%. Decrease: The membership degree is 0.3, and the value range is [-10%, -5%], and the representative value is -7.5%.

[0086] Since the centroid method is a defuzzification method in fuzzy control, its principle is based on the idea of moment balance, and the centroid position of the fuzzy set is used as the exact value after defuzzification. Suppose in a discrete fuzzy set, there are fuzzy elements , and their corresponding membership degrees are . Understanding from a physical sense, each fuzzy element can be regarded as a mass point with mass , located at position . Then, the centroid position of the entire fuzzy set is equivalent to the point of action of the resultant force of these mass points. According to the principle of moment balance, assuming the calculation formula for the centroid position can be expressed as: ; where is the value corresponding to the centroid position, is the representative value of the th element, is the membership degree of the th element, and is the number of fuzzy elements in the fuzzy set. Based on this formula, the centroid position (i.e., the motor speed adjustment value) corresponding to the second fuzzy set regarding the motor speed adjustment value in the above example is, .

[0087] Further optionally, although the traditional centroid method has good effects on symmetric membership functions (such as triangles and trapezoids), when the membership function is asymmetric or multi-modal, it is vulnerable to extreme values. In response to this situation, the above formula can be further improved. For example, a weight adjustment factor can be introduced, and this weight adjustment factor can be set according to the physical meaning of the fuzzy elements or the system safety level. Based on the calculation formula for the above centroid position , the calculation formula for the improved centroid position is: ; where is the weight adjustment factor, and the meanings of other parameters are the same, so they will not be elaborated here. For example, in the adjustment of the tower crane motor speed, set to 0.5 for "substantially accelerating" (high-risk state), and set to 1 for "slightly adjusting", reducing the weight of dangerous operations. In this way, through the weight adjustment factor, the influence of unreasonable extreme values can be suppressed, and the control safety can be improved.

[0088] Further optionally, to adapt to the control requirements of different operation stages of the tower crane, the system introduces a phased dynamic configuration mechanism for the weight adjustment factor. When the tower crane switches from the "heavy load hoisting stage" to the "precision positioning stage", this mechanism can intelligently adjust the control strategy: during the hoisting stage, rapid movement is required, and the system increases the weight of the fuzzy elements related to "speed adjustment" (such as setting it to 1.2) and decreases the weight of "position accuracy" (such as setting it to 0.8), making the center of gravity calculation tend to efficient movement; during the positioning stage, high-precision fine-tuning is required, so the weights are adjusted in the opposite direction, the weight of "speed adjustment" is reduced to 0.6, and the weight of the elements related to "position deviation" is increased to 1.5, enabling the calculation results to meet the requirements of fine control. The system determines the stage conversion by real-time monitoring of the hook height, load sway and other states, and dynamically adjusts the influence of each fuzzy element on the center of gravity calculation, avoiding the control imbalance of traditional fixed weights in different stages, effectively improving the hoisting efficiency by 20% and reducing the positioning error by 30%, and solving the problem that a single weight is difficult to adapt to the complex control requirements of multiple stages.

[0089] The maximum membership degree method is to find the element with the maximum membership degree in the fuzzy set and take its corresponding exact value as the defuzzification result. If there are multiple elements with the maximum membership degree, the average value, minimum value or maximum value can be taken. This method is simple to calculate and the result is fast, which is very suitable for the tower crane control system with high real-time requirements and can give control parameters in time according to the fuzzy reasoning result. For example, when the tower crane makes an emergency obstacle avoidance or quickly adjusts its position, the parameters can be quickly determined; when there is a dominant element in the fuzzy reasoning (the membership degree of a certain state is much higher than others), the exact value of this element can be directly used as the result to determine the parameters. In scenarios where high requirements are placed on the intuitive simplicity of the results, such as operator training or simple operations, directly selecting representative elements is convenient for understanding and executing the parameters.

[0090] In an example of a calculation process of the maximum membership degree method, assume that the fuzzy elements regarding the braking force of the braking device in the second fuzzy set are "strong", "medium", and "weak", and their membership degrees and value ranges are as follows: Strong: the membership degree is 0.2, and the value range is [80, 100] (assuming a certain quantization unit of the braking force), and the midpoint value is 90. Medium: the membership degree is 0.5, and the value range is [50, 70], and the midpoint value is 60. Weak: the membership degree is 0.3, and the value range is [0, 30], and the midpoint value is 15. Here, the membership degree of "medium" is the largest, so according to the maximum membership degree method, the midpoint 60 of the value range of the fuzzy element "medium" is selected as the adjustment value of the braking force of the braking device. In this way, the control parameters of the device can be quickly determined according to the fuzzy reasoning result. In scenarios with high real-time requirements, the tower crane equipment can respond to operation instructions in time, quickly perform mode switching and equipment control, and improve the operation efficiency.

[0091] The above embodiments combine the centroid method and the maximum membership degree method to determine the tower crane control parameters. The centroid method calculates the weighted average of the fuzzy set elements and outputs precise parameters; the maximum membership degree method takes the midpoint of the range corresponding to the element with the maximum membership degree, and the result is intuitive and the calculation is fast. According to the characteristics of the parameter data, the two complement each other's advantages: the centroid method is used for high-precision operations to ensure precise control, such as lifting precision components; the maximum membership degree method is used in conventional fast-switching scenarios to improve the response speed, taking into account both precision and real-time performance, effectively enhancing the control performance of the tower crane and ensuring the safety and efficiency of the operation.

[0092] Step S105: Display the tower crane task model to the user.

[0093] Step S106: After the user confirms the tower crane task model, send the tower crane task model to the control system to control the tower crane equipment to complete the corresponding operation tasks according to the operation time sequence.

[0094] In the tower crane assisted driving system, in step S105, the task model is converted into a visual interface. The operation nodes are displayed with a time axis or a flow chart (such as "lifting preparation → load detection → horizontal movement → precise placement → return to the original position and stop"), and each node is marked with the execution order, the estimated time consumption, and the dependency relationship (such as "horizontal movement" needs to be triggered after "the lifting height reaches the standard"), and is dynamically associated with the real-time device status (such as the boom angle and the hook position). If the detected data exceeds the safety threshold, it will be marked in red for warning. Based on the digital twin model of the construction site integrating BIM coordinates and environmental data, the visual hoisting path and spatial constraints can be scaled and rotated to view the details. For example, the planned trajectory is highlighted in blue, the obstacles (such as adjacent tower cranes and buildings) are marked with a red semi-transparent model, and the safety distance (such as no entry within 2 meters) is dynamically rendered with a yellow warning area. By zooming and rotating to view the path details, for example, at the "precise placement" node, after zooming in, the millimeter-level precision requirements of the target position and the corresponding camera vision guidance range can be seen. In addition, it also supports the user to manually adjust the task parameters or insert temporary nodes, and real-time verify the feasibility of the adjusted ones and update the risk assessment results.

[0095] Taking the display device as a virtual reality (VR) device as an example, it is suitable for complex working conditions (such as high-altitude heavy-load lifting). Through the immersive 3D scene, the operator seems to be in the tower crane operation space. The operator can view the node details through gesture interaction (such as grabbing the "horizontal movement" node to display the angular velocity limit of the boom at this stage, 0.5° / s), or view the spatial relationship between the hook and the target position in real time by turning the head.

[0096] As an alternative embodiment, in step S105, the tower crane task model is constructed as a visualization model; the visualization model is displayed through a linkage console or virtual reality device. In the embodiment of the present application, multiple task nodes associated with the to-be-executed operation tasks are sequentially displayed in the visualization model in the order of operation time sequence from the earliest to the latest. In step S105 of the tower crane assisted driving system, by constructing a visualization task model and interactively displaying it with the aid of a linkage console or VR device, the operation process is made transparent and operable. The model construction adopts time-sequential modeling of task nodes, decomposes the tower crane operation into multi-level nodes, encapsulates the basic attributes, execution parameters and safety information to ensure the complete presentation of the operation logic and safety strategy; at the same time, uses data-driven rendering technology to synchronize the real-time status collected by sensors, compares and displays the planned and actual trajectories, and constructs a 3D scene based on BIM or CAD, marking the paths and obstacles. In terms of display adaptation, the linkage console displays the task nodes in a 2D structured interface of an industrial touch screen, presents the risk level and planned duration with a Gantt chart, supports clicking to view parameters, dragging to adjust the task order and automatically verifying the feasibility. The VR device brings an immersive experience, arranges the task nodes with floating cubes, combines the real scene with the virtual model, supports gesture operations to view details, and is convenient for the operator to perceive the deviation between the hook and the target position.

[0097] Exemplarily, taking the hoisting of bridge components at the construction site of a high-speed railway pier as an example, the operator views the task model through the VR device. The "ground inspection" node displays the wire rope wear degree and load weight in real time. The "heavy load hoisting" node highlights the hoisting path to avoid ground vehicles and automatically switches the control mode. The "high-altitude precise docking" node embeds a visual guidance interface to feedback the hook offset in real time. The linkage console synchronously displays the Gantt chart, highlighting the wind speed and the safety threshold of load sway of the "high-altitude precise docking" node, and prompts the parameter adjustment of the anti-wind mode in real time when the wind speed changes.

[0098] By means of model-driven display and multi-modal interaction, the abstract task planning is transformed into an intuitive interface that can be perceived and intervened. The time-sequential node display enables the operator to clearly master the operation process and key parameters, reducing misoperations caused by information asymmetry.

[0099] Further optionally, it is assumed that each task node is provided with a control for managing or viewing the to-be-executed operation task. Based on the above assumption, after presenting the tower crane task model to the user in step S105, it further includes: modifying the steps, order, and task type in the to-be-executed operation task corresponding to the task node by modifying the control; deleting the to-be-executed operation task corresponding to the task node by deleting the control; adjusting the device control parameters in the to-be-executed operation task corresponding to the task node by the parameter adjustment control; reconstructing the tower crane task model based on the maintained tower crane task model; generating a visualization model corresponding to the tower crane task model and pushing it to the user side to confirm whether to execute the reconstructed tower crane task model.

[0100] Specifically, in the tower crane assisted driving system, when each task node is provided with a control for managing or viewing the tasks to be executed, after the user views the visual task model in step S105, the task nodes can be dynamically maintained through the interactive controls.

[0101] For example, in the 2D interface of the linkage console or the 3D scene of the VR device, a gear-shaped modification control, a trash can-shaped deletion control, and a slider-type parameter adjustment control are provided beside each task node: Clicking the modification control can pop up a hierarchical menu, allowing the user to drag and adjust the task steps (such as moving the "load detection" node before "lifting preparation"), change the task order (such as splitting the "horizontal movement" node into two sub-nodes of "left shift to avoid obstacles → right shift for positioning" due to obstacles), or modify the task type (such as adjusting "ordinary lifting" to "precision positioning" and activating the corresponding macro-control strategy). Clicking the deletion control can remove redundant nodes (such as deleting the preset "wind resistance mode verification" node on a sunny day) and automatically verify the integrity of the task logic after deletion (such as confirming whether the remaining nodes meet the load safety threshold). The sliding parameter adjustment control can modify the device control parameters in real time (such as fine-tuning the motor speed of the "precision placement" node from 0.1 m / s to 0.08 m / s, or relaxing the allowable range of position deviation to ±1.5 cm), and the interface synchronously displays the impact of the parameter changes on the task duration and safety margin during the adjustment (such as the reduction in speed resulting in an increase in duration of 2 minutes and an increase in safety margin of 15%).

[0102] After the user completes the maintenance operation, based on the updated task steps, sequence, type, and parameters, re-parse the matching relationship between the operation instructions and the real-time job information. For example, when the task type changes from "ordinary lifting" to "precision positioning", automatically load a higher-precision vision servo algorithm and PID control parameters, and trigger multi-modal data fusion verification (such as re-planning the path by combining the hook position sensor and visual camera data). The re-constructed tower crane task model will generate a new visualization interface, highlighting the changed parts in the form of a Gantt chart on the linkage console (such as the adjusted nodes are marked with an orange border), and prompting the change of the task sequence with a flashing cube in the VR device. At the same time, push a confirmation pop-up window containing the modification summary (such as "Add an obstacle avoidance node, estimated time increase of 5 minutes"), parameter adjustment details (such as "Motor speed reduced by 20%, accuracy improved to ±1 cm"), and risk assessment results (such as "Collision risk reduced from 35% to 12%") for the user to confirm again. If the user confirms to execute, generate a control instruction stream based on the maintained model. If there are logical conflicts (such as deleting a key safety verification node), the pop-up window will mark the error reason (such as "Load detection node missing, unable to verify the overload risk") and lock the modification to ensure the safety and feasibility of the task model. This dynamic maintenance mechanism with control interaction enables the operator to flexibly adjust the operation plan according to the real-time working conditions. For example, quickly insert an obstacle avoidance step when a new obstacle is found, or temporarily modify the control parameters when the equipment status is abnormal. Through real-time verification and visual feedback to form a closed loop, it not only retains the efficiency of intelligent planning but also gives the flexibility of manual intervention, significantly improving the task adaptation ability and operation safety of the tower crane in complex construction sites.

[0103] In step S106, after the user confirms the tower crane task model, first perform two-way verification to ensure the reliability and safety of the task. On the one hand, verify the timing logic of the task nodes to ensure that action nodes such as "lower" are triggered after the preconditions such as "above the target position" are met, and at the same time check the matching of the control strategy and the equipment capabilities (such as the "micro-manipulation" node is only activated when the hook speed <0.1 m / s). On the other hand, based on the tower crane dynamics model and real-time environmental parameters (such as wind speed, load weight), calculate the safety margin of each node (such as the ratio of the load weight to the rated load, the difference between the obstacle distance and the minimum safety distance). If the verification fails (such as the margin <20%), automatically return to the user interface and highlight the risk nodes (such as the "horizontal movement" node of "heavy load + close obstacle"), prompting the user to correct.

[0104] Further optionally, in step S106 of the tower crane assisted driving system, the verified task model is converted into a control system instruction stream through the industrial protocol adaptation layer. The graph structure encoding technology is adopted to disassemble the task nodes into hardware control signals and software policy calls under the binary protocol. For example, the task model in graph structure (including node and edge relationships) is encoded into binary industrial protocols such as Profinet and EtherCAT, and each task node is disassembled into underlying control signals. For example, "lifting" corresponds to the motor operation instruction, and "accurate placement" triggers the vision servo module. At the same time, a priority control sequence is generated according to the risk level and time constraint, the interruption permission is given to the safety instruction, and the instruction is encrypted and transmitted through digital signature to ensure the safety and reliability of the instruction. After receiving the instruction, intelligent execution is realized with the help of the task scheduler: the corresponding control module is activated according to the operation timing, the progress is dynamically monitored in combination with the sensor data, and the task rhythm is automatically adjusted in case of environmental changes. For example, when executing the "horizontal movement" node, the path planning algorithm is called, and when executing the "wind resistance mode" node, the pre-trained dynamic compensation parameters are loaded, and the execution progress is dynamically monitored through the data such as the hook position and the boom angle transmitted back by the sensor in real time. For example, in case of environmental changes such as a sudden increase in wind speed, the "horizontal movement" speed is automatically reduced and the time window of the subsequent node is extended to ensure the adaptive adjustment of the task rhythm. When risks such as obstacle intrusion and load sway exceeding the threshold are detected, the "safe retreat" mechanism is immediately triggered, the operation is paused, the safety mode is switched, and the exception details are pushed. After the risk is eliminated, the operation is resumed in an orderly manner, forming a closed-loop control of "planning - execution - feedback - adjustment". Further, through the visual interaction interface, the task confirmation time of the operator is shortened and dynamic adjustment is supported. The two-way verification mechanism can reduce the logical error rate and avoid most of the safety risks. The real-time synchronization of the plan and the actual state can be applied to complex working conditions such as "heavy load + strong wind", improve the operation success rate, and realize the intelligence, high efficiency and high safety of tower crane operation.

[0105] In the embodiment of the present application, through the adaptive decision-making model, real-time mode configuration and tower crane task model, the intelligent control of the tower crane equipment is realized, which helps to improve the adaptability of the operator to complex working conditions, reduce human errors and ensure operation safety. At the same time, with the help of a variety of assisted driving modes, the operation difficulty can be reduced and the equipment control efficiency can be improved. In particular, the adaptive decision-making model has the ability of autonomous learning, and can improve the tower crane control efficiency through the integration of a variety of assisted driving modes and intelligent decision-making mechanisms under complex working conditions, and comprehensively improve the operation efficiency, accuracy and safety.

[0106] Please refer to Figure 2 , Figure 2A tower crane control system 200 based on assisted driving provided by an embodiment of the present application. The tower crane control system 200 based on assisted driving includes a deployment module 201, a decision-making module 202, a configuration module 203, a task module 204, a display module 205, and a distribution module 206. The deployment module 201 is configured to deploy a control system in a tower crane device; the control system is connected to a linkage console, and a variety of assisted driving modes are integrated in the control system; the variety of assisted driving modes are respectively used to assist a user in implementing control operations on the tower crane device by using corresponding operation modes; the variety of assisted driving modes at least include: a linkage console mode, a voice driving mode, a fixed-distance driving mode, a macro-control mode, and a somatosensory control mode; the decision-making module 202 is configured to construct an adaptive decision-making model of the tower crane device by using an A3C algorithm according to historical state information of the tower crane device, historical mode selection information, and historical execution conditions of tower crane operation tasks, so as to select the assisted driving mode most matching the tower crane device; the configuration module 203 is configured to obtain real-time state information of the tower crane device, and configure a target mode matching the real-time state information by using the adaptive decision-making model; the task module 204 is configured to receive an operation instruction sent by a user to the tower crane device, and construct a tower crane task model of the tower crane device in the target mode based on the operation instruction; a plurality of task nodes are sequentially arranged in the tower crane task model in the order from first to last according to the operation time sequence, and each task node corresponds to a to-be-executed operation task; the operation instruction is at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a somatosensory control instruction; the display module 205 is configured to display the tower crane task model to the user; the distribution module 206 is configured to, after the user confirms the tower crane task model, distribute the tower crane task model to the control system, so as to control the tower crane device to complete corresponding operation tasks according to the operation time sequence. In some embodiments, the tower crane control system 200 based on assisted driving can be applied to a terminal device. It should be noted that, for the sake of convenience and brevity of description, the specific working process of the tower crane control system 200 based on assisted driving described above can refer to the corresponding process in the foregoing embodiment of the tower crane control method based on assisted driving, and will not be elaborated herein.

[0107] Please refer to Figure 3 , Figure 3 A schematic block diagram of the structure of a terminal device provided by an embodiment of the present application.

[0108] As Figure 3As shown in the figure, the terminal device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a bus 303, which is, for example, an I2C bus. Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit, and it can also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc. Specifically, the memory 302 can be a Flash chip, a read-only memory disk, an optical disc, a USB flash drive, or a mobile hard disk, etc.

[0109] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of some structures related to the solution of the embodiment of the present application, and does not constitute a limitation on the terminal device to which the solution of the embodiment of the present application is applied. Specifically, the server may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. Among them, the processor is used to run the computer program stored in the memory and implement any one of the tower crane control methods based on assisted driving provided by the embodiment of the present application when executing the computer program. It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described terminal device can refer to the embodiment of the tower crane control method based on assisted driving described above, and will not be repeated here.

[0110] The embodiment of the present application also provides a storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the tower crane control methods based on assisted driving provided in the specification of the embodiment of the present application.

Claims

1. A tower crane control method based on assisted driving, characterized in that, The method includes: Deploying a control system in a tower crane device; the control system is connected to a linkage console, and multiple assisted driving modes are integrated in the control system; the multiple assisted driving modes are respectively used to assist the user to implement control operations on the tower crane device by using corresponding operation modes; the multiple assisted driving modes at least include: a linkage console mode, a voice driving mode, a fixed-distance driving mode, a macro-control mode, and a somatosensory control mode; According to the historical state information of the tower crane device, the historical mode selection information, and the historical execution situation of the tower crane operation task, an adaptive decision-making model of the tower crane device is constructed by using the A3C algorithm to select the assisted driving mode that best matches the tower crane device; Obtaining the real-time state information of the tower crane device and configuring the target mode that matches the real-time state information by using the adaptive decision-making model; Receiving an operation instruction issued by the user to the tower crane device, and constructing a tower crane task model of the tower crane device in the target mode based on the operation instruction; multiple task nodes are sequentially arranged in the tower crane task model in the order from first to last according to the operation time sequence, and each task node corresponds to a to-be-executed operation task; the operation instruction is at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a somatosensory control instruction; Displaying the tower crane task model to the user; after the user confirms the tower crane task model, sending the tower crane task model to the control system to control the tower crane device to complete the corresponding operation task according to the operation time sequence; Wherein, after receiving the operation instruction issued by the user to the tower crane device, it further includes: parsing the operation instruction to obtain the operation task requirements in the operation instruction; judging whether the current target mode of the tower crane device meets the operation task requirements; if the target mode does not meet the operation task requirements, reselecting the target mode based on the operation instruction; inputting the operation instruction, the real-time operation information of the tower crane device, and the reselected target mode into a fuzzy control model to generate device control parameters during the target mode switching process for smoothly switching the tower crane device from the current target mode to the reselected target mode; the device control parameters at least include: a motor speed adjustment value, a braking force of a braking device, and a luffing mechanism speed; Among them, the real-time operation information at least includes: the position of the boom of the tower crane equipment, the position of the hook, the equipment motion state information, the load information, and the environmental information; the operation instruction, the real-time operation information of the tower crane equipment, and the re-selected target mode are input into the fuzzy control model to generate the equipment control parameters during the target mode switching process, so as to smoothly switch the tower crane equipment from the current target mode to the re-selected target mode, including: taking the operation target, operation type, the real-time operation information, and the re-selected target mode in the operation instruction as input variables, identifying the value range and actual physical meaning of each input variable, and configuring a corresponding first fuzzy set and membership function for each input variable based on the identification result; calculating the membership degree of each input variable in its corresponding first fuzzy set through the membership function; based on the membership degree and the preset fuzzy rules, performing fuzzy inference on the target mode switching process of the tower crane equipment to obtain a second fuzzy set for constructing the equipment control parameters; performing defuzzification processing based on the second fuzzy set to obtain the equipment control parameters.

2. The tower crane control method based on assisted driving according to claim 1, wherein, The performing defuzzification processing based on the second fuzzy set to obtain the equipment control parameters includes: Adopting the centroid method, determining the centroid position of the second fuzzy set according to the membership degree and value range of each fuzzy element in the second fuzzy set, and taking the value corresponding to the centroid position as the first equipment control parameter; Adopting the maximum membership degree method, selecting the fuzzy element with the maximum membership degree according to the membership degree and value range of each fuzzy element in the second fuzzy set, and taking the midpoint of the value range of the selected fuzzy element as the second equipment control parameter; Based on the data value characteristics of the equipment control parameters, selecting the finally output equipment control parameter from the first equipment control parameter and the second equipment control parameter.

3. The tower crane control method based on assisted driving according to claim 1, characterized in that, The constructing an adaptive decision-making model of the tower crane equipment by using the A3C algorithm according to the historical state information of the tower crane equipment, the historical mode selection information, and the historical execution situation of the tower crane operation task to select the auxiliary driving mode most suitable for the tower crane equipment includes: Defining multiple auxiliary driving modes as an action space, and selecting an auxiliary driving mode from the action space at each time step; Setting a reward function for the adaptive decision-making model based on the historical execution situation; the reward function at least includes: task completion reward, completion time reward, completion quality reward, and abnormal situation penalty of the tower crane equipment; Obtaining the historical state information of the tower crane equipment, the historical mode selection information, and the historical execution situation of the tower crane operation task; Constructing multiple operation task environments according to the historical state information; Inputting the historical state information and the historical mode selection information into multiple operation task environments to parallelly train the Actor network for selecting the auxiliary driving mode; Inputting the historical state information, the historical mode selection information, and the historical execution situation into multiple operation task environments to parallelly use the Critic network for evaluating the value of the Actor network; Update the network parameters of the Actor network and the Critic network using the said reward function to obtain an adaptive decision-making model for the tower crane equipment.

4. The tower crane control method based on assisted driving according to claim 3, wherein The configuration of the target mode matching the real-time state information using the said adaptive decision-making model includes: Obtain the image data around the tower crane equipment, and perform obstacle positioning based on the image data to obtain the obstacle positioning data of the tower crane equipment; Parse the real-time state information to obtain the real-time equipment component state, real-time weather condition, and real-time wind speed condition of the tower crane equipment; Input the real-time equipment component state, the real-time weather condition, the real-time wind speed condition, and the obstacle positioning data into the adaptive decision-making model, use the Actor network to predict the driving mode, and obtain the prediction probabilities of the tower crane equipment in each assisted driving mode; randomly select a candidate mode through roulette based on the prediction probabilities; Monitor the safety risks of the tower crane equipment in the candidate mode; if the safety risk is greater than the set threshold, set the target mode to the assisted driving mode matching the safety risk item, or switch the target mode to the assisted driving mode with a safety level higher than the current target mode.

5. The tower crane control method based on assisted driving according to claim 1, wherein The construction of the tower crane task model for the tower crane equipment based on the operation instruction includes: Select a target parsing model matching the operation instruction from the candidate algorithm model library; Process the operation instruction using the target parsing model to obtain an operation instruction feature vector; the operation instruction feature vector contains the instruction sequence between each operation task, the task type to which the operation task belongs, and the logical relationship; Based on the operation instruction feature vector and the pre-set association relationship between the task type and the task priority, determine the operation tasks to be executed and the arrangement method of the operation tasks; According to the operation tasks to be executed and the arrangement method of the operation tasks, combine them through a graph neural network to obtain the tower crane task model.

6. The tower crane control method based on assisted driving according to claim 1, wherein The display of the tower crane task model to the user includes: Construct the tower crane task model into a visualization model; Display the visualization model through a linkage console or virtual reality device; the visualization model sequentially displays multiple task nodes associated with the operation tasks to be executed in chronological order of operation; Each task node is provided with a control for managing or viewing the operation tasks to be executed; after displaying the tower crane task model to the user, it further includes: Modify the steps, order, and task type in the operation task to be executed corresponding to the task node by modifying the control; Delete the operation task to be executed corresponding to the task node by deleting the control; Adjust the equipment control parameters in the operation task to be executed corresponding to the task node by adjusting the parameter control; Re-construct the tower crane task model based on the maintained tower crane task model; generate a visualization model corresponding to the tower crane task model and push it to the user side to confirm whether to execute the re-constructed tower crane task model.

7. A tower crane control system based on assisted driving, characterized in that, The system includes: Deployment module, used to deploy a control system in a tower crane device; the control system is connected to a linkage console, and multiple auxiliary driving modes are integrated in the control system; the multiple auxiliary driving modes are respectively used to assist the user to control the tower crane device by adopting corresponding operation modes; the multiple auxiliary driving modes at least include: linkage console mode, voice driving mode, fixed-distance driving mode, macro-control mode, somatosensory control mode; Decision-making module, used to construct an adaptive decision-making model of the tower crane device by using the A3C algorithm according to the historical state information, historical mode selection information of the tower crane device, and the historical execution situation of the tower crane operation task, so as to select the auxiliary driving mode that best matches the tower crane device; Configuration module, used to obtain the real-time state information of the tower crane device, and configure the target mode that matches the real-time state information by using the adaptive decision-making model; Task module, used to receive the operation instruction issued by the user to the tower crane device, and construct a tower crane task model of the tower crane device in the target mode based on the operation instruction; multiple task nodes are sequentially arranged in the tower crane task model in the order from first to last according to the operation time sequence, and each task node corresponds to a to-be-executed operation task; the operation instruction is at least one of a linkage console operation instruction, a voice instruction, a fixed-distance driving instruction, a macro-control instruction, and a somatosensory control instruction; Display module, used to display the tower crane task model to the user; Issuing module, used to, after the user confirms the tower crane task model, issue the tower crane task model to the control system to control the tower crane device to complete the corresponding operation task according to the operation time sequence; Among them, after receiving the operation instruction issued by the user to the tower crane device, the task module is further used to: analyze the operation instruction to obtain the operation task requirements in the operation instruction; judge whether the current target mode of the tower crane device meets the operation task requirements; if the target mode does not meet the operation task requirements, reselect the target mode based on the operation instruction; input the operation instruction, the real-time operation information of the tower crane device, and the reselected target mode into a fuzzy control model to generate equipment control parameters during the target mode switching process, so as to smoothly switch the tower crane device from the current target mode to the reselected target mode; the equipment control parameters at least include: motor speed adjustment value, braking device braking force, luffing mechanism speed; Among them, the real-time operation information at least includes: the position of the boom of the tower crane device, the position of the hook, the equipment motion state information, the load information, and the environment information; The task module inputs the operation instruction, the real-time operation information of the tower crane equipment, and the reselected target mode into the fuzzy control model to generate equipment control parameters during the target mode switching process. When the tower crane equipment is smoothly switched from the current target mode to the reselected target mode, it is specifically used for: taking the operation target, operation type, the real-time operation information, and the reselected target mode in the operation instruction as input variables, identifying the value range and actual physical meaning of each input variable, and configuring a corresponding first fuzzy set and membership function for each input variable based on the identification result; calculating the membership degree of each input variable in its corresponding first fuzzy set through the membership function; performing fuzzy inference on the target mode switching process of the tower crane equipment based on the membership degree and the preset fuzzy rules to obtain a second fuzzy set for constructing the equipment control parameters; and performing defuzzification processing based on the second fuzzy set to obtain the equipment control parameters.

8. A terminal device, characterized in that, The terminal device includes a processor and a memory; The memory is used for storing computer programs; The processor is used for executing the computer program and implementing the tower crane control method based on assisted driving according to any one of claims 1 to 6 when executing the computer program.

Images