All-domain safety monitoring and intelligent early warning system and method for construction site tower crane operation

Abstract

The invention relates to the technical field of safety monitoring and early warning, and discloses a global safety monitoring and intelligent early warning system and method for construction site tower crane operation, and the method comprises the steps: deploying a multi-layer sensor network at a plurality of target parts of a construction site tower crane, and collecting the operation state data of the tower crane; carrying out extended Kalman filtering processing on the tower crane operation state data to obtain a tower crane sensor installation angle parameter and a non-integrity constraint lever arm parameter; performing global state estimation based on the tower crane sensor mounting angle parameter and the non-integrity constraint lever arm parameter to obtain a tower crane global state and external disturbance data; inputting the global state of the tower crane and the external disturbance data into a deep neural network, and performing safety boundary calculation to obtain a dynamic safety boundary model of the construction site tower crane; and performing risk measurement calculation based on the dynamic safety boundary model to obtain a safety risk grading early warning instruction, thereby improving the safety and collaboration of construction site tower crane operation.

Description

Technical Field

[0001] This application relates to the technical field of safety monitoring and early warning, and particularly to a global safety monitoring and intelligent early warning system and method for tower crane operations at construction sites. Background Art

[0002] Tower cranes at construction sites, as important vertical transportation equipment, are widely used in various construction projects. However, tower crane operations face safety hazards, especially in the complex construction site environment with multiple tower cranes operating in coordination. Traditional tower crane safety monitoring systems mainly rely on single sensors or simple monitoring measures, making it difficult to cope with complex and changing working conditions. Especially when tower cranes face unknown external disturbances such as wind loads and load fluctuations, their safety monitoring and early warning capabilities are significantly insufficient. This technical limitation has led to frequent occurrence of tower crane operation safety accidents, causing heavy casualties and property losses.

[0003] A key problem with existing tower crane monitoring systems is that individual tower cranes cannot obtain global information. In the coordinated operation of multiple tower cranes, the status information between tower cranes cannot be effectively shared and integrated, resulting in inaccurate assessment of the overall safety situation. At the same time, the problem of sensor installation errors has been long ignored, causing inaccurate monitoring of the motion state and affecting the reliability of early warning decisions. In addition, traditional monitoring systems generally have a problem of response delay. Especially when facing emergencies, they cannot provide a fast enough response and early warning. The early warning time is highly correlated with the initial state and lacks the guarantee of convergence within a fixed time, resulting in insufficient timeliness of early warning. Summary of the Invention

[0004] This application provides a global safety monitoring and intelligent early warning system and method for tower crane operations at construction sites, thereby improving the safety and coordination of tower crane operations at construction sites.

[0005] In the first aspect of this application, a global safety monitoring and intelligent early warning method for tower crane operations at construction sites is provided. The global safety monitoring and intelligent early warning method for tower crane operations at construction sites includes: Deploy a multi-level sensor network at multiple target parts of the tower crane at the construction site and collect tower crane operation status data; Perform extended Kalman filter processing on the tower crane operation status data to obtain tower crane sensor installation angle parameters and non-integrity constraint boom parameters; Based on the tower crane sensor installation angle parameters and the non-integrity constraint boom parameters, perform global state estimation to obtain the tower crane global state and external disturbance data; Input the tower crane global state and the external disturbance data into a deep neural network to perform safety boundary calculation to obtain the dynamic safety boundary model of the tower crane at the construction site; Based on the dynamic safety boundary model, perform risk metric calculation to obtain a safety risk classification early warning instruction.

[0006] The second aspect of the present application provides a global safety monitoring and intelligent early warning system for tower crane operations at construction sites. The global safety monitoring and intelligent early warning system for tower crane operations at construction sites includes: A deployment module, configured to deploy a multi-level sensor network at multiple target locations of a tower crane at a construction site and collect tower crane operation state data; A filtering processing module, configured to perform extended Kalman filtering processing on the tower crane operation state data to obtain tower crane sensor installation angle parameters and non-integrity constraint lever arm parameters; A state estimation module, configured to perform global state estimation based on the tower crane sensor installation angle parameters and the non-integrity constraint lever arm parameters to obtain tower crane global state and external disturbance data; A safety boundary calculation module, configured to input the tower crane global state and the external disturbance data into a deep neural network to perform safety boundary calculation to obtain a dynamic safety boundary model of the tower crane at the construction site; A risk metric calculation module, configured to perform risk metric calculation based on the dynamic safety boundary model to obtain a safety risk classification early warning instruction.

[0007] Compared with the prior art, the present application has the following beneficial effects: The multi-level sensor network deployment and data acquisition technology realizes the all-round and multi-dimensional acquisition of tower crane operation status data. By arranging various types of sensors on the tower crane boom, slewing mechanism, end part, counterweight part, and construction site environment, combined with industrial-grade wireless transmission and data preprocessing technology, the comprehensiveness and reliability of data acquisition are ensured, providing a high-quality data basis for subsequent state estimation and risk analysis. The tower crane state parameter estimation method based on non-integrity constraints solves the problem of inaccurate state estimation caused by sensor installation errors in traditional monitoring systems. Through the establishment of non-integrity constraint equations and extended Kalman filter processing, the accurate estimation of sensor installation angle parameters and non-integrity constraint lever arm parameters is realized, significantly improving the accuracy of tower crane motion trajectory prediction and laying a foundation for safety monitoring. The design of the distributed fixed-time observer overcomes the disadvantages of slow convergence speed and dependence on the initial state of traditional observation methods, enabling each tower crane to quickly estimate the spatial state parameters required for safe operation under the condition that most individuals cannot obtain the global state, while accurately estimating the external disturbances suffered by each tower crane. The upper bound of the early warning response time is not affected by the initial state and only depends on the eigenvalues of the information transfer matrix and the monitor parameters. The dynamic safety boundary model based on deep learning breaks through the limitations of traditional fixed-threshold monitoring. Through the design of a multi-layer neural network structure and a specific loss function, the dynamic adjustment of the safety domain is realized, and the safety boundary can be adaptively generated according to the current tower crane state and external environment changes, improving the accuracy and environmental adaptability of safety monitoring. The combination of the non-singular distributed fixed-time controller and the risk metric calculation method realizes the accurate quantification and hierarchical early warning of tower crane operation risks. By designing a reasonable risk metric function and a hierarchical early warning mechanism, corresponding early warning and control measures are taken for different risk levels, thereby minimizing safety risks to the greatest extent while ensuring the operation efficiency of the tower crane. The integrated application of the digital twin model and the global collaborative processing technology realizes the global collaborative monitoring of the tower crane group on the construction site. Through functions such as high-precision three-dimensional modeling, collision detection, and trajectory planning, combined with the multi-agent reinforcement learning algorithm, the optimal multi-tower collision avoidance collaborative strategy is generated, greatly improving the safety and collaboration of tower crane operations on the construction site. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0009] The structures, proportions, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those skilled in this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they do not have substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0010] Figure 1 is a schematic flowchart of the method for global safety monitoring and intelligent early warning of tower crane operations at the construction site provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the structure of the system for global safety monitoring and intelligent early warning of tower crane operations at the construction site provided by an embodiment of the present invention. Detailed implementation manners

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] The flowchart shown in the drawings is only an example for illustration, and does not necessarily include all the content and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0013] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0014] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the method for global safety monitoring and intelligent early warning of tower crane operations at the construction site in the embodiments of this application includes: Step 100: Deploy a multi-level sensor network at multiple target parts of the tower crane at the construction site and collect the tower crane operation status data; It can be understood that the execution entity of this application can be the global safety monitoring and intelligent early warning system for the operation of tower cranes at construction sites, or it can also be a terminal or a server, and specific details are not limited here. In this embodiment of the application, the server is taken as an example of the execution entity for illustration.

[0015] Specifically, high-precision strain sensors are installed on the boom of the construction site tower crane. These sensors are evenly distributed at key positions on the boom. For example, the strain sensors are installed at intervals of one-fifth of the total length of the boom to ensure that the force conditions of the entire boom are accurately sensed. This layout can monitor the strain changes of the boom under different loads and wind loads and provide sufficient redundant data to cope with the situation of individual sensor failures or data loss. Angle sensors and MEMS sensors are installed at the slewing mechanism. The former is used to provide accurate rotation angle data, and the latter can provide auxiliary angular velocity information for the system at low cost. The combination of high- and low-precision sensors can effectively improve the measurement accuracy through data fusion technology and reduce the system's dependence on high-cost sensors. At the same time, triaxial acceleration sensors are installed at the end of the tower crane boom and the counterweight end to collect three-dimensional acceleration data in real time and effectively capture the dynamic motion characteristics of the end of the tower crane boom, especially the vibration and swing amplitude generated at the end when the tower crane starts, brakes, and suddenly encounters wind loads. Meteorological stations need to be installed at the four corners of the entire construction site. These meteorological stations are equipped with wind speed, wind direction, temperature, and humidity sensors to monitor environmental parameters in real time at a frequency of 1 Hz. A load sensor is installed in the driver's cab to monitor the actual load condition of the tower crane in real time. The accuracy reaches 0.01 tons, and the measurement range covers 0 to 100 tons, which can effectively prevent overloading operations and predict the operating safety of the tower crane by analyzing the load change trend. The parameters of the strain sensors, angle sensors, acceleration sensors, meteorological stations, and load sensors are configured. For example, the strain sensors are configured with appropriate sensitivity coefficients, and the angle sensors and MEMS sensors are set with corresponding sampling frequencies (such as 50 Hz). The acceleration sensors need to adjust their measurement ranges (±16 g) and sampling frequencies (100 Hz). The sensors of the meteorological station ensure that their parameters are within the standard range, and the load sensor needs to be calibrated to ensure accuracy. The parameter configuration process of these sensors is completed through a dedicated software interface or an embedded system, and the configuration data is uniformly uploaded to the data management platform to ensure that all acquisition devices are in an ideal working state. When the sensors are started, the collected raw data is transmitted in real time through an industrial-grade wireless transmission module. To ensure the security of data transmission, the AES-256 encryption algorithm is used to encrypt the data. In each cycle of wireless signal propagation (10 ms), the data is encrypted, packaged, and sent to the receiving terminal to obtain the first operating state data stream. During the data transmission process, the integrity of the data stream is checked to ensure the reliability of data transmission in a high-interference environment.The received first operating state data stream enters the data preprocessing stage. The wavelet transform is used to denoise the original data. The wavelet transform method has good local characteristics in both the time-frequency domain, which can effectively separate the noise components in the sensor signal, thereby retaining the key signal features. On this basis, the Kalman filter algorithm is used to smooth the denoised data. The Kalman filter makes the data stream maintain stability while dynamically changing through a prediction-correction cycle process, obtaining the second operating state data stream. The second operating state data stream is compressed. The compression algorithm uses a specific data compression method, such as the compression technology based on feature point extraction, and controls the data compression ratio within the range of 10:1. This compression method can maximize the retention of the key features of the data and avoid losing important information due to compression. After completing the data compression, the compressed data is stored in the distributed database according to the unified timestamp and data format. The unified timestamp enables the data of different sensors to be accurately aligned on the time line, and the consistent data format ensures the compatibility and operability of the data in the subsequent analysis process. The tower crane operating state data is obtained.

[0016] Step 200: Perform extended Kalman filter processing on the tower crane operating state data to obtain the tower crane sensor installation angle parameters and the non-integrity constraint boom parameters; Specifically, the configuration vectors of the tower crane's operating state data are extracted. The configuration vectors are used to describe the specific position and attitude of the tower crane in three-dimensional space, including the position information of the tower crane's boom, the angular state of the slewing mechanism, the movement trajectory of the boom tip, and the dynamic data collected by various sensors (such as acceleration sensors, angle sensors, and strain sensors). After obtaining the configuration vectors, a nonholonomic constraint-based model of the tower crane is established. This model defines the constraint relationships of the system through the geometric characteristics and physical motion laws of the tower crane. For example, when the tower crane rotates and the boom moves, the position of the boom tip is not only affected by the movement of the boom itself but also restricted by the dynamic characteristics of the entire tower crane structure. This restrictive relationship is modeled through geometric and kinematic relationship models. Based on the nonholonomic constraint-based model, the installation error matrix of the sensors is calculated to eliminate the measurement inaccuracies caused by position offsets and installation angle errors during the installation of the sensors. For example, sensors installed on the tower crane boom may have angle errors due to installation tilt or loose fixation. The calculation of the error matrix needs to consider the angular deviations of the sensors in three dimensions and convert these deviations into correctable values through a reasonable error model. At the same time, the sensor installation error matrix is combined with the three-dimensional position data of the tower crane boom to construct a state estimator, enabling the system to correct the state estimation deviation caused by installation errors in real time during operation. The prediction equation and observation equation of the state estimator are constructed. The prediction model predicts the state at the next moment based on the historical data of the tower crane's movement and known motion laws, while the observation model uses the actual sensor data to correct the prediction result. For example, when the prediction model estimates that the boom tip of the tower crane is at a certain specific position, the observation model uses the data of the acceleration sensor, angle sensor, and strain sensor to verify this prediction. If there is a deviation between the actual observed data and the predicted data, the estimated value is dynamically adjusted through a filter to gradually approach the true state. In the prediction-correction cycle of the extended Kalman filter, the high-precision dynamic tracking of the tower crane's operating state is achieved by continuously updating the state estimation value. An adaptive adjustment strategy is implemented based on the extended Kalman filter parameter model. By dynamically updating the process noise covariance and observation noise covariance, the automatic adjustment of the filter parameters is realized. When the operating state of the tower crane changes drastically (such as a sudden increase in wind load or a sharp change in load), the adaptive strategy identifies the change trend in the data and automatically increases the weight of the model for the new data, enabling the filter to quickly adapt to the changing environment. When the operating state of the tower crane is relatively stable, the error range is automatically tightened to improve the stability and accuracy of state estimation. The adaptive extended Kalman filter is iteratively calculated. Each iteration compares the error between the current estimated value and the actual observed value, and minimizes the estimation error by adjusting the Kalman gain.In this process, there are two stopping conditions for iterative calculation: one is that the change rate of the estimated parameters is less than a preset threshold, for example, the change rate is less than a very small angular unit to ensure that the estimated value has tended to be stable; the other is to force the stop when the number of iterations reaches the preset maximum number to prevent the system from falling into an infinite iterative calculation state. Through the iterative calculation of the extended Kalman filter, the installation angle parameters of the tower crane sensors and the non-integrity constraint boom parameters are obtained.

[0017] Step 300: Based on the installation angle parameters of the tower crane sensors and the non-integrity constraint boom parameters, perform global state estimation to obtain the global state of the tower crane and external disturbance data; It should be noted that the multi-tower crane system at the construction site is modeled as a network system with N nodes. In this network model, each tower crane is regarded as an independent node, and the communication connection between nodes is represented by network edges. This network structure can reflect the physical position relationship between tower cranes and define the communication topology between tower cranes through the relationship between nodes and edges. For example, multiple tower cranes achieve data sharing and status synchronization through industrial Ethernet, wireless sensor networks or dedicated communication protocols (such as CAN bus, Modbus). This networked communication method enables each tower crane to obtain the status data of adjacent tower cranes in real time. Based on the tower crane network communication model and the tower crane sensor installation angle parameters, the state equation of the tower crane at the construction site is constructed to obtain the dynamic system model of the tower crane. The state equation is used to describe the dynamic behavior of the tower crane system under different operating conditions, including the position information, speed, angle, load change of the tower crane and the influence of external disturbances on the system. The dynamic system model not only considers the motion state of a single tower crane, but also couples the states of multiple tower cranes through the network communication model to realize the collaborative calculation of the states of multiple tower cranes. For example, when a tower crane swings significantly, through the communication model, the adjacent tower cranes can obtain this information in real time and dynamically adjust their own operation modes to avoid potential collision risks. According to the tower crane dynamic system model and the non-integrity constraint boom parameters, a distributed fixed-time observer structure is designed, so as to converge to the true state within a finite time without being affected by the initial state. During the design process of the observer structure, the sensor installation angle parameters and the non-integrity constraint boom parameters are introduced. By embedding these parameters into the state equation, the observer dynamically adjusts the estimation strategy during the actual operation process, making the estimation result more in line with the actual working conditions. For example, when there is an error in the sensor installation angle, the observer continuously corrects the state estimation process and introduces the error into the calculation model, making the finally output global state more accurate and reliable. The gain matrix of the distributed fixed-time observer is designed to obtain the optimal observer gain matrix. The role of the gain matrix is to adjust the response speed and accuracy of the observer to the change of the system state. By setting different gain coefficients, the "forgetting" speed of the observer to historical data when receiving new data is controlled. For example, when the system detects that the external disturbance is large, the gain matrix automatically increases the weight of the new data, enabling the observer to quickly adapt to the new state; while when the system state is relatively stable, the gain coefficient is reduced to enhance the stability and anti-noise ability of the system. In order to enhance the estimation accuracy of external disturbances, based on the observer gain matrix, the immersion and invariance theory is used to design a finite-time dynamic scale disturbance observer to obtain the disturbance observer. The design goal of the disturbance observer is to capture the external disturbances received by the system, including factors such as wind load changes, load fluctuations, and mechanical vibrations that affect the operation stability of the tower crane. Through the immersion and invariance theory, the disturbance observer separates the external disturbance signal from the sensor data within a short time and distinguishes these disturbance signals from the normal operation state of the system.For example, when the system detects abnormal fluctuations in the acceleration sensor data, the disturbance observer analyzes the pattern of data changes to determine whether this change is due to an increase in external wind load or improper operation of the tower crane. Based on the distributed fixed-time observer and the disturbance observer, the upper bound of the estimated time is calculated and iteratively calculated within a fixed time. The purpose of calculating the upper bound of the estimated time is to ensure that in the worst case, the system can also complete state estimation and disturbance identification within the specified time, avoiding potential safety hazards caused by calculation delays. By introducing a fixed-time convergence algorithm in the observer design, the estimation error is guaranteed to converge in each iteration, and the final global state and external disturbance data are output within a fixed time.

[0018] Step 400: Input the global state of the tower crane and the external disturbance data into the deep neural network for safety boundary calculation to obtain the dynamic safety boundary model of the tower crane at the construction site; Specifically, a set of key parameters for the safe operation of the tower crane is determined based on the global state of the tower crane and external disturbance data, including various parameters that affect safety during the operation of the tower crane, such as boom strain value, slewing angular acceleration, load rate, wind speed, wind direction, acceleration, angular velocity, and environmental temperature and humidity. A safety domain is defined for each parameter, and these safety domains limit the safe range of the parameters through lower and upper limit values. By defining the safety domains of each parameter, a multi-dimensional safety domain is obtained. A deep neural network structure is constructed based on the multi-dimensional safety domain to achieve the prediction and calculation of the dynamic safety boundary. When designing the deep neural network, the input layer is set to contain e neurons, and these neurons respectively correspond to the global state of the tower crane and external disturbance parameters, ensuring that all key factors affecting the safe operation of the tower crane can be perceived and processed by the model. The hidden layer of the neural network consists of four layers of neurons, and the number of neurons in each layer is 256, 128, 64, and 32 in sequence, ensuring the learning ability of the model and achieving the gradual refinement of features by gradually reducing the number of neurons layer by layer, thereby enhancing the model's recognition ability for complex data patterns. The output layer is designed with 2m neurons, where m represents the number of key parameters, and the 2m neurons respectively output the upper and lower limits of the safety domain of each parameter, enabling the neural network to simultaneously predict the safe ranges of all key parameters and form an adaptive safety boundary model. In the selection of the activation function, the Leaky ReLU is used in the hidden layer. This activation function still retains a certain gradient when the input is less than zero, thus avoiding the problem of "neuron death" and improving the training stability of the model. Data augmentation processing is performed based on the global state of the tower crane and external disturbance data to construct the training dataset of the neural network. Based on the global state of the tower crane and external disturbance data, data augmentation techniques are used to effectively expand the training sample set and enhance the generalization ability of the model. For example, by adding Gaussian noise, random scaling, random translation, and rotation to the data, the model can cope with the randomness in the data and environmental changes. And the sampling balance technique is used to solve the problem of uneven sample distribution in the dataset, especially giving higher weights to those rare samples in dangerous states, thereby ensuring that the model has stronger sensitivity to dangerous states when predicting the safety boundary. After obtaining the augmented training dataset, a loss function is designed for the deep neural network. This loss function includes the traditional mean square error loss and introduces a constraint loss and a regularization term. The constraint loss is used to ensure that the predicted safety domain always conforms to physical and engineering constraint conditions. For example, when the wind speed is too high, the safe operation window is automatically reduced, while the regularization term prevents the model from overfitting by introducing a penalty when the model overly relies on specific training samples, thereby improving the performance of the model on new data. These loss terms are combined into the training objective function of the neural network through weighted coefficients to ensure that the model training process is always optimized in the direction of improving the prediction accuracy of the safety boundary.During the training process of the neural network model, the training dataset and the training objective function are input into the Adam optimizer. The Adam optimizer is an adaptive learning rate optimization algorithm that improves the model convergence speed by dynamically adjusting the learning rate and maintains a stable learning efficiency when the model approaches the optimal solution. To enhance the training effect of the model, a cosine annealing learning rate strategy is adopted. This strategy gradually reduces the learning rate as the number of iterations increases during the training process. When the model approaches the global optimal solution, fine-tuning is performed with a smaller learning rate to avoid the risk of skipping the optimal solution, and a trained deep neural network model is obtained. The global state of the tower crane and the external disturbance data are input into the trained deep neural network model. The model calculates the upper and lower limits of the safety domain of m key parameters through forward propagation, and these upper and lower limit values constitute an adaptive safety boundary. When the state parameters of the tower crane during actual operation approach or exceed these safety boundaries, the system issues a warning signal in real time to remind the operator to take appropriate safety measures through visual, acoustic, or automatic control means. A dynamic safety boundary model driven by a deep neural network is obtained.

[0019] Step 500: Perform risk metric calculation based on the dynamic safety boundary model to obtain a safety risk classification warning instruction.

[0020] Specifically, based on the dynamic safety boundary model, a single-machine risk metric function for construction site tower cranes is designed to quantify the gap between the current operating state of the tower crane and the safety boundary. The risk metric function is achieved by calculating the distance between the key state parameters of the tower crane and the preset safety domain. For example, when parameters such as the boom strain, wind speed, and load of the tower crane approach their safety upper limits, the risk metric function will give a higher risk value, indicating that the tower crane is in a dangerous working state, while when these parameters are within the safe range, the risk value is lower. Through this risk metric function, the safety of each key parameter of the tower crane is monitored in real time. Based on the single-machine risk metric function, a global risk metric function for multiple construction site tower cranes is constructed. The operating environment of multiple tower cranes is more complex, and the interaction between each tower crane and the shared working space pose safety threats to each other. When constructing the global risk metric function, not only the individual state of each tower crane is considered, but also the spatial relationship between tower cranes and the interference factors of the external environment are integrated. The global risk metric function integrates the safety states of multiple tower cranes through a weighted formula, enabling the safety of tower crane operations across the entire construction site to be uniformly evaluated. For example, when two tower cranes are operating in the same area, if the distance between them is too close, even if the single-machine risk metric values of each tower crane are within the safe range, the global risk metric value may still be high, indicating a collision risk between the tower cranes. Based on the global risk metric function, a non-singular distributed fixed-time controller is designed. The design goal of this controller is to ensure that the tower crane can stably reach the predetermined safety state within a fixed time during operation and can quickly adjust when encountering disturbances. The design of the non-singular distributed fixed-time controller involves the optimization of control strategies and gain matrices. The role of the gain matrix is to adjust the speed and accuracy of the controller's response to changes in the current state of the tower crane. The design of the controller's gain matrix needs to be adjusted according to the motion characteristics of the tower crane, the environmental factors of the construction site, and the interaction between multiple tower cranes. By optimizing the gain matrix, precise control of the tower crane operation is achieved, ensuring that the tower crane always remains in a safe state in a dynamic environment. According to the global risk metric function, the risk levels are divided into four levels, and a risk classification standard is established. These risk levels are: safety level, attention level, warning level, and danger level. The division of each level is based on the output value of the global risk metric function. For example, when the risk metric value is lower than a certain threshold, the operation of the tower crane is at the safety level and no early warning needs to be triggered; when the risk metric value is in the medium range, the tower crane is at the attention level, and the system will send a visual prompt to the operator to remind them of potential risks; when the risk metric value approaches or exceeds the warning threshold, the tower crane will enter the warning level, triggering an audible and visual alarm and restricting the operation speed of the tower crane to reduce potential safety risks; when the risk metric value reaches the highest threshold, the tower crane enters the danger level, immediately triggering an emergency brake and locking the movement of the tower crane to prevent accidents. Through this classification mechanism, corresponding early warning measures are taken according to different risk levels, thus ensuring the safety of tower crane operations.According to the set risk classification criteria and the non-singular distributed fixed-time controller, corresponding early warning measures are formulated for different risk levels. At the safe level, the system does not trigger any early warnings to ensure that the tower crane operation is not interfered; at the attention level, the system sends visual cues to the operator, such as flashing indicator lights or warning messages on the display screen, to remind the operator to pay attention to the current working environment and state changes; at the warning level, audible and visual alarms are issued, and the operating speed of the tower crane is also restricted to reduce the risks during operation and ensure that the tower crane can respond in a timely manner when encountering potential dangers; at the danger level, an emergency braking mechanism is triggered to quickly stop the operation of the tower crane through the automatic control system and lock the movement of the tower crane to avoid major accidents.

[0021] A three-dimensional space coordinate system is established according to the actual layout of the construction site and the geometric parameters of the tower crane. The three-dimensional point cloud data of the buildings and equipment on the construction site is obtained through laser scanning technology. The scanning technology uses lidar equipment, which captures the three-dimensional contours and spatial information of the buildings, tower cranes and surrounding equipment with extremely high precision, and obtains a high-precision 3D map of the construction site. Based on the obtained high-precision 3D map and the existing CAD model, a tower crane structure model, a building model and an obstacle model are constructed to obtain a complete three-dimensional model of the construction site environment. In this process, the tower crane structure model needs to reflect all the movable parts of the tower crane, including the boom, slewing mechanism, luffing mechanism and hoisting system, etc., and correspond the motion characteristics of these parts to the actual physical parameters, such as the length of the tower crane boom, the rotation angle range, the hoisting speed, etc. At the same time, the building model not only includes the external contour of the building, but also contains important internal structure information, such as elements that may affect the safety of tower crane operation, such as windows, platforms, external pipelines, etc. The obstacle model is mainly used to describe dynamic obstacles such as temporary facilities, stacked materials, and moving equipment on the construction site. These models need to be connected to real-time data sources (such as Internet of Things devices or cameras) to dynamically update their positions and states to ensure that the models are synchronized with the actual scenario. Kinematic and dynamic modeling of the tower crane is carried out on the three-dimensional model of the construction site environment to accurately simulate the motion behavior of the tower crane in three-dimensional space. In the kinematic model, the motions of the three degrees of freedom of the tower crane's slewing, luffing and hoisting are described. By establishing motion equations, the telescoping, rotation of the tower crane boom and the lifting and lowering of the hook are simulated. In the dynamic modeling, the forces and torques acting on the tower crane during these motion processes are calculated. For example, when the wind load or load changes, the force conditions of each part of the tower crane structure and the resulting dynamic responses are calculated. By combining kinematic and dynamic parameters, a tower crane kinematic and dynamic model is obtained. The global state of the tower crane, the safety risk classification and early warning instructions, and the tower crane kinematic and dynamic model are integrated into the digital twin platform to construct a hierarchical construction site tower crane digital twin model. The hierarchical architecture of this platform includes a data layer, a model layer, a service layer and an application layer. The data layer collects, stores and processes real-time data from sensors and external data sources. The model layer contains physical models, behavior models and prediction models. The service layer provides core algorithms for data analysis, state estimation and risk assessment. The application layer is oriented to end users and realizes functions such as visual monitoring, early warning prompts and decision support. In the digital twin platform, the actual state of the tower crane and the virtual model are synchronized in real time. When the tower crane moves during actual operation, the motion trajectory and operation state of the tower crane are simultaneously presented in the digital twin model, thus realizing a seamless connection between the physical world and the digital world. Based on the digital twin model, a collision detection module and a trajectory planning module are designed to ensure the safety during multi-tower crane collaborative operation.The collision detection module uses the GJK algorithm to calculate the minimum distance and collision risk between tower cranes in real time. By calculating the nearest distance between polygons or polyhedra, this algorithm can efficiently determine whether a tower crane is in potential collision danger in a complex three-dimensional space. At the same time, the trajectory planning module comprehensively considers obstacle constraints and dynamic constraints through the A* algorithm and the dynamic window method to plan a safe and efficient movement trajectory for the tower crane. For example, when the system detects that two tower cranes are approaching each other during operation, the trajectory planning module will calculate the optimal avoidance path and avoid collisions while maintaining the operation efficiency. According to the tower crane collision risk assessment and safety trajectory planning scheme, a multi-agent reinforcement learning algorithm is used to generate a collaborative risk avoidance strategy. Each tower crane is regarded as an agent, and these agents gradually optimize their operation strategies by learning historical data and simulated operations, so that in a complex construction site environment, even in the presence of dynamic obstacles and external disturbances, multiple tower cranes can automatically cooperate, maintain a safe distance and complete tasks efficiently.

[0022] In the embodiments of the present application, the omnidirectional acquisition of tower crane operation state data is realized through a multi-level sensor network, providing a high-quality data basis for state estimation; the sensor installation error problem is solved based on the non-integrity constraint model and the extended Kalman filtering technology, improving the accuracy of motion trajectory prediction; the distributed fixed-time observer enables each tower crane to quickly estimate the spatial state parameters and external disturbances under the condition of being unable to obtain the global state, and the upper bound of the early warning response time is not affected by the initial state; the deep learning dynamic safety boundary model breaks through the limitations of traditional fixed-threshold monitoring and realizes the adaptive generation of the safety domain; the combination of the non-singular distributed fixed-time controller and the risk metric calculation method realizes accurate risk quantification and hierarchical early warning; the integrated application of digital twin technology and global collaborative processing, through high-precision three-dimensional modeling and multi-agent reinforcement learning algorithm, generates the optimal multi-tower crane collision avoidance collaborative strategy, significantly improving the safety and collaboration of tower crane operations at the construction site.

[0023] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Install strain sensors at the boom part of the tower crane at the construction site, install angle sensors and MEMS sensors at the slewing mechanism, install three-axis acceleration sensors at the end of the boom and the end of the counterweight, install weather stations at the four corners of the construction site, and install load sensors in the driver's cab to obtain a multi-level sensor network structure; Configure the parameters of the strain sensors, angle sensors, acceleration sensors, weather stations and load sensors, and collect the original operation state data based on the multi-level sensor network; Transmit the original operating status data through an industrial-grade wireless transmission module to obtain the first operating status data stream, and perform wavelet transform denoising processing and Kalman filtering processing on the first operating status data stream to obtain the second operating status data stream; Perform compression processing on the second operating status data stream to obtain compressed data, and store the compressed data in a distributed database according to a unified timestamp and data format to obtain the tower crane operating status data.

[0024] Specifically, a variety of sensors are installed at the key parts of the tower crane to monitor the operating status of the tower crane. At the boom part of the tower crane, four strain sensors are installed, and the spacing distance between the sensors is 1 / 5 of the total length of the boom to ensure that the entire stressed area of the tower crane boom is covered. When the tower crane is working, the boom will bear the pressure from various external factors such as load and wind. The strain sensors collect the minute deformations on the boom in real time. By analyzing these strain data, it is judged whether the boom is in an overloaded or unsafe working state, and an alarm is issued in a timely manner. At the slewing mechanism of the tower crane, a high-precision angle sensor and a low-precision MEMS sensor are installed to monitor the angle change and angular velocity change during the slewing process of the tower crane. The high-precision angle sensor provides accurate slewing angle information, while the low-precision MEMS sensor captures the subtle changes in angular velocity. By monitoring the slewing angle and angular velocity in real time, it is ensured that the operation of the tower crane is within the safe range. The sampling frequency of the sensors is set to 50Hz to obtain slewing data with sufficient accuracy and speed. At the end of the boom and the end of the counterweight of the tower crane, triaxial acceleration sensors are installed respectively, and the measurement range is set to ±16g, and the sampling frequency is set to 100Hz. The acceleration sensors can capture the dynamic response of the tower crane under factors such as load fluctuations and wind changes. Especially when the tower crane is hoisting or slewing, there will be acceleration or deceleration phenomena. Through the triaxial acceleration data obtained in real time by the acceleration sensors, the working status of the tower crane is monitored, and potential safety hazards are identified, such as the severe vibration of the tower crane when the wind force or load is too large. At the four corners of the construction site, weather stations are installed to monitor environmental parameters such as wind speed, wind direction, temperature and humidity in real time, and the sampling frequency is set to 1Hz. Environmental conditions (such as wind speed) directly affect the stability of the tower crane. In the tower crane driver's cab, a load sensor is installed, with a measurement range of 0 to 100 tons and an accuracy of 0.01 tons. The load sensor monitors the load condition of the tower crane in real time to judge whether the tower crane is overloaded. Overloading will increase the risk of tower crane failures. By monitoring the load in real time and comparing it with the preset safe load limit, it is ensured that the tower crane is always within the safe working range. The original operating status data is collected in real time. These data are transmitted to the industrial-grade wireless transmission module and are securely transmitted through AES-256 encryption. The wireless transmission module transmits various types of data collected in real time to the data processing center through a 10ms cycle, forming the first operating status data stream. To ensure the accuracy and reliability of the data, the first operating status data stream is denoised. The denoising process is completed through wavelet transform. Wavelet transform decomposes the data into low-frequency and high-frequency components, thereby removing high-frequency noise. After wavelet transform, the data is smoothed by a Kalman filter. The Kalman filter corrects the data through a prediction-update method to improve the estimation accuracy. Through continuous iteration, the Kalman filter provides a more accurate state estimate and reduces the influence of external interference. After wavelet transform denoising and Kalman filter processing, the second operating status data stream is obtained. The second operating status data stream is compressed to reduce the storage and transmission burden.The compression uses a feature point extraction method to control the data compression ratio within the range of 10:1, ensuring that key information is not lost. The compressed data is stored in a distributed database according to a unified timestamp and format to ensure data consistency and efficient storage, obtaining the tower crane operation status data. In a specific embodiment, the process of executing step 200 may specifically include the following steps: Extract the configuration vector from the tower crane operation status data and establish a basic model of the non-integrity constraints of the tower crane; Calculate the sensor installation error matrix according to the basic model of the non-integrity constraints of the tower crane, and construct a state estimator based on the sensor installation error matrix and the three-dimensional position data of the tower crane boom; Construct the prediction equation and the observation equation for the state estimator to obtain the extended Kalman filter parameter model; Implement an adaptive adjustment strategy based on the extended Kalman filter parameter model, update the process noise covariance and the observation noise covariance to obtain an adaptive extended Kalman filter; Perform iterative calculations on the adaptive extended Kalman filter, and stop the iteration when the change rate of the estimated parameters is less than the preset threshold or the number of iterations reaches M times, obtaining the tower crane sensor installation angle parameters and the non-integrity constraint rod arm parameters.

[0025] Specifically, extract the configuration vector from the operation data of the tower crane, which includes the position information and motion state of each component of the tower crane. The configuration vector of the tower crane is expressed as , which includes all variables related to the structure and motion of the tower crane, such as the position, angle, rotation speed, etc. of the boom. The configuration vector of the tower crane is expressed as a combination of multiple sub-vectors, such as:

[0026] Among them, represents the position coordinates of a certain part of the tower crane, is the rotation angle of this part. Similarly, other sub-vectors represent the relevant information of other parts of the tower crane. This configuration vector is the basis for describing the entire operation state of the tower crane and is constructed through data collected by sensors (such as angle sensors, acceleration sensors, etc.). Establish a basic model of the non-integrity constraints of the tower crane. The non-integrity constraint model describes the mutual constraint relationship between each part of the tower crane and is constructed based on the geometric and dynamic characteristics of the tower crane. During the operation of the tower crane, parts such as the boom and slewing mechanism will move under different constraint conditions, and these constraint conditions are represented by the following non-integrity equations:

[0027] Among them, is a constraint matrix related to the geometry of the tower crane, which includes the constraint conditions of the kinematics of the tower crane, is the time derivative of the configuration vector, representing the motion speed of the tower crane. Through this constraint equation, it describes how the tower crane is stressed and moves under different states, ensuring the coordination and stability of each part of the tower crane. Calculate the sensor installation error matrix according to the nonholonomic constraint basic model of the tower crane. The sensor installation error matrix is calculated by analyzing the installation angle error of each sensor. Assuming the installation error of the sensor is (the installation angle errors around the X, Y, and Z axes respectively), the sensor error matrix is expressed in the form of a rotation matrix:

[0028] where, is the rotation matrix around the X axis, is the rotation matrix around the Y axis, is the rotation matrix around the Z axis. Through the above rotation matrices, the installation error of the sensor is converted into an error matrix , which is used to correct the error of the sensor output. Construct a state estimator based on the sensor installation error matrix and the three-dimensional position data of the tower crane boom. The state estimator estimates the true state of the tower crane through the data of the sensor, especially in the case where the sensor has errors. The state estimator uses filtering techniques, such as the extended Kalman filter, to handle state estimation in nonlinear systems. Assuming the state vector of the tower crane is , the core of the state estimator is to update the state estimation using the difference between the observed value and the predicted value . To achieve this goal, construct the prediction equation and the observation equation for the state estimator to obtain the parameter model of the extended Kalman filter. The prediction equation of the extended Kalman filter predicts the state at the next moment based on the estimation of the current state of the tower crane:

[0029] where, is the state estimation at the current moment, is the control input, is the state transition function, which describes the transformation from the current state to the next state. The observation equation maps the state vector to the actual observation space and is expressed as:

[0030] where, is the observed value at the current moment, is the observation model of the state vector, is the observation noise. The extended Kalman filter updates the state iteratively, using the prediction equation and the observation equation to calculate the current state estimate. To improve the accuracy of the state estimate, an adaptive adjustment strategy is implemented based on the extended Kalman filter parameter model to update the process noise covariance and the observation noise covariance, resulting in an adaptive extended Kalman filter. The operating environment and load conditions of the tower crane are constantly changing, so the covariances of the process noise and the observation noise also change over time. To cope with these changes, an adaptive adjustment strategy is used to dynamically update the process noise covariance and the observation noise covariance . The update formula takes the following form:

[0031]

[0032] where and are forgetting factors used to adjust the weights of historical data, set between 0.95 and 0.99. Through these update rules, the sensitivity of the state estimator to environmental changes is ensured, so as to maintain a high accuracy under changing working conditions. The adaptive extended Kalman filter is iteratively calculated, and the iteration stops when the change rate of the estimated parameter is less than the preset threshold or the number of iterations reaches M times, obtaining the installation angle parameters of the tower crane sensors and the non-integrity constraint boom parameters. The specific stopping condition is set as:

[0033] where is the preset threshold, indicating that when the change amount of the state estimate is less than this value, it is considered to have converged, is the maximum number of iterations. After the iteration is completed, the finally obtained contains the installation angle parameters of the tower crane sensors and the non-integrity constraint boom parameters.

[0034] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Model the tower cranes at multiple construction sites as a network system with N nodes and construct a tower crane network communication model; Based on the tower crane network communication model and the installation angle parameters of the tower crane sensors, construct a state equation of the tower crane at the construction site to obtain a tower crane dynamic system model; According to the tower crane dynamic system model and the non-integrity constraint boom parameters, design a distributed fixed-time observer structure to obtain a distributed fixed-time observer; Design the gain matrix of the distributed fixed-time observer to obtain the observer gain matrix; Based on the observer gain matrix, the immersion and invariance theory is used to design a finite-time dynamic scale disturbance observer, and the disturbance observer is obtained. According to the distributed fixed-time observer and the disturbance observer, calculate the upper bound of the estimated time, and iteratively calculate within a fixed time to obtain the global state of the tower crane and the external disturbance data.

[0035] Specifically, the multi-tower crane system at the construction site is modeled as a network system with nodes, and a tower crane network communication model is constructed. In this model, the tower cranes at the construction site are regarded as a multi-node network, each tower crane is regarded as a node in the network, and the communication connection between tower cranes represents an edge in the network. The topological graph of the network is defined as a graph containing nodes, and the node set represents the tower cranes at the construction site, and the edge set

[0036] represents the communication relationship between tower cranes. By constructing the communication topology of the tower crane system, the communication behavior and collaborative operation between tower cranes are simulated. The state equation of the tower crane network is expressed as: where, represents the state vector of the th tower crane, including key parameters such as position, velocity, angle, angular velocity, etc.; and are known non-linear functions that describe the dynamic behavior of the tower crane and the influence of the control input respectively; is the control input of the tower crane (such as the motion control of the tower crane boom); ; where, is the state estimate of the th tower crane, is the estimate of the external disturbance, is the set of tower cranes adjacent to the tower crane , is the communication weight between the tower crane and the tower crane , and are gain matrices that respectively adjust the state difference between tower cranes and the influence of external disturbances on the observer. The gain matrix Designed as a diagonal matrix, the calculation formula for its diagonal elements is:

[0037] where, and are the largest eigenvalue and the second smallest eigenvalue of the Laplacian matrix respectively. The Laplacian matrix is an important matrix describing the topological structure of the tower crane network, and its eigenvalues are related to the connectivity of the network. The gain matrix is designed as:

[0038] where, , this design ensures that the observer converges quickly and stably under external disturbances. The immersion and invariance theory is used to design a finite-time dynamic scaling disturbance observer to estimate the external disturbance. The design formula of the disturbance observer is:

[0039] where, is the error between the estimated state and the actual state, is a positive definite diagonal gain matrix representing the gain of the disturbance estimation, is a design parameter, and is selected to ensure the convergence speed and accuracy. The disturbance observer compensates for the external disturbance of the tower crane, enabling the system to accurately estimate the disturbance within a fixed time. To ensure the convergence of the system within a fixed time, an upper bound of the estimation time is designed to calculate the convergence time of the tower crane system. According to the immersion and invariance theory, the upper bound of the estimation time is expressed as:

[0040] where, is the eigenvalue ratio of the Laplacian matrix . Through this formula, it is ensured that each tower crane estimates the global state and external disturbance within a fixed time, thus avoiding the problems of slow global state estimation speed and low accuracy in traditional tower crane monitoring systems. According to the distributed fixed-time observer and disturbance observer, iterative calculations are performed, state estimation is carried out at each moment, and dynamic adjustment is made according to the changes of the system. When the change rate of the estimated parameter is less than the preset threshold or the number of iterations reaches the maximum value, the iteration stops, and the global state estimation and external disturbance data of the tower crane are output.

[0041] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Determine a set of key parameters for the safe operation of the tower crane based on the global state of the tower crane and external disturbance data, and define a safety domain for each parameter in the set of key parameters to obtain a multi-dimensional safety domain; Construct a deep neural network structure based on the multi-dimensional safety domain. The input layer of the deep neural network structure contains e neurons corresponding to the tower crane state and external disturbance parameters. The hidden layer contains 4 layers, and the number of neurons in each layer is 256, 128, 64, and 32 respectively. The output layer contains 2m neurons corresponding to the upper and lower limits of the safety domain of m parameters; Perform data augmentation processing based on the global state of the tower crane and external disturbance data to obtain a neural network training data set, and design a loss function for the deep neural network structure to obtain a neural network training objective function; Input the neural network training data set and the neural network training objective function into the Adam optimizer to execute the cosine annealing learning rate strategy to obtain a trained deep neural network model; Input the global state of the tower crane and external disturbance data into the trained deep neural network model, output the upper and lower limits of the safety domain of m parameters, form an adaptive safety boundary, and obtain a dynamic safety boundary model for the tower crane on the construction site.

[0042] Specifically, determining a set of key parameters for the safe operation of the tower crane based on the global state of the tower crane and external disturbance data includes parameters such as the boom strain value, slewing angular acceleration, load rate, and wind speed of the tower crane. Assume that the set of key parameters for the safe operation of the tower crane is , each corresponds to a key tower crane operation parameter, such as load, wind speed, boom strain, etc. These parameters directly affect the safety of the tower crane. For each parameter , define a safety domain , where is the lower limit of the parameter , is its upper limit. Define a safety domain for each parameter in the set of key parameters to construct a multi-dimensional safety domain , which represents the safe operation space of the tower crane on all key parameters. In order to dynamically predict the safety boundary of the tower crane, a dynamic safety boundary model of the tower crane is established based on a deep neural network. The deep neural network structure is designed as follows: the input layer contains neurons, representing the state of the tower crane and external disturbance parameters. These input parameters include real-time state data such as the load, wind speed, and acceleration of the tower crane. The hidden layer of the network contains four layers, and the number of neurons in each layer is 256, 128, 64, and 32 respectively. The activation function is selected as Leaky RelU. LeakyReLU can avoid the problem of gradient disappearance when the input value is negative and improve the convergence speed of the model during training. The formula of Leaky ReLU is as follows:

[0043] Among them, is set to a small value (e.g., 0.01) for linear transformation of negative input values. The output layer contains neurons, corresponding to the upper and lower limits of the safety domain of each parameter respectively. For each parameter , the output layer will predict and , that is, the lower and upper limits of the parameter, and these predicted values represent the safety boundaries under the current tower crane state. To improve the generalization ability and robustness of the model, data augmentation technology is adopted to expand the training samples. The data augmentation method generates new training samples by adding Gaussian noise, random scaling, and random offset to the original data, and the augmentation ratio is set to 1:5. This means that each training sample generates five variants through data augmentation, effectively increasing the diversity of training samples and improving the robustness of the neural network. At the same time, to accelerate the training and improve the performance of the neural network, batch normalization technology is introduced. Batch normalization helps to reduce the internal covariate shift and improve the stability and speed of network training. Batch normalization standardizes the input of each layer so that the input data has the same distribution, avoiding the problems of gradient explosion or gradient disappearance. To train this deep neural network, a comprehensive loss function is designed, which contains multiple parts. The loss function is expressed as:

[0044] Among them, is the mean squared error loss, which measures the gap between the predicted upper and lower limits of the safety domain and the true values; is the constraint loss to ensure that the safety domain output by the model conforms to the physical constraint conditions (such as the operation limits of the tower crane); is the regularization term, which is used to prevent overfitting. , and are weight coefficients, which are used to adjust the proportion of each loss term in the total loss, and are set to , 0.5, . During the training process, the Adam optimizer is used to optimize the parameters of the model. The Adam optimizer accelerates the training process by adaptively adjusting the learning rate, and its update rule is as follows:

[0045]

[0046] ;

[0047] Among them, is the learning rate, and is the exponential decay rate, usually set to 0.9 and 0.999. The Adam optimizer dynamically adjusts the learning rate to avoid the problem of the learning rate being too large or too small at the beginning of training. In addition, to address the problem of imbalanced data (i.e., fewer rare but dangerous state data), the Focal Loss is adopted to increase higher weights for rare state samples. The Focal Loss formula is:

[0048] where, is the probability of the predicted class, is the balancing factor, is the modulating factor, usually set to 2, which is used to amplify the loss of difficult-to-classify samples. By using the Focal Loss, the network can better identify rare dangerous states and avoid underestimating the impact of these important samples. The cosine annealing learning rate strategy is adopted during the training process. The initial learning rate is set to 0.001 and gradually decreases during the training process. It is achieved through the following formula:

[0049] where, is the initial learning rate, is the current iteration number, is the total number of training epochs. The cosine annealing learning rate strategy helps the network converge to the optimal solution in the later stage of training. After the training is completed, the trained neural network model is input with the global state of the tower crane and external disturbance data, and the upper and lower limits of the safety domain of each parameter are output, so as to form an adaptive safety boundary and obtain the dynamic safety boundary model of the tower crane on the construction site.

[0050] In a specific embodiment, the process of executing step 500 may specifically include the following steps: Design a single-machine risk metric function for the tower crane on the construction site based on the dynamic safety boundary model, and construct a global risk metric function for multiple tower cranes on the construction site according to the single-machine risk metric function; Design a nonsingular distributed fixed-time controller based on the dynamic safety boundary model, and design the gain matrix of the nonsingular distributed fixed-time controller to obtain the controller gain matrix; Divide the risk level into four levels according to the global risk metric function of multiple tower cranes on the construction site to obtain the risk classification standard; Formulate warning measures for different risk levels according to the risk classification standard and the nonsingular distributed fixed-time controller. The safety level does not trigger a warning, the attention level sends a visual prompt to the operator, the warning level triggers an audible and visual alarm and restricts the operating speed of the tower crane, and the danger level triggers an emergency brake and locks the movement of the tower crane to obtain the safety risk classification warning instruction.

[0051] Specifically, design a safety risk metric function for the tower crane system. The risk metric function for each tower crane reflects the relationship between the state of the tower crane and its safety domain. Assume that the th parameter of the tower crane corresponds to the th safety domain , with a weight coefficient of . Then the risk metric function for this tower crane is:

[0052] where, is the number of tower crane parameters, is the parameter to its safety domain . The distance function is defined based on the relationship between the parameter and its safety domain, and the formula is as follows:

[0053] This formula means that when the parameter falls within the safety domain , the risk is 0; if the parameter exceeds the safety domain, the above formula calculates the relative distance between the parameter and the safety domain boundary, thus reflecting the risk of tower crane operation. Define the system global risk metric function. The global risk metric of the multi-tower crane system is the maximum value of all tower crane risk metrics, indicating the overall safety condition on the construction site. The global risk metric function is:

[0054] where, is the total number of tower cranes, is the risk metric function of the th tower crane. By calculating the risk metrics of all tower cranes and taking the maximum value, the maximum risk metric of the entire construction site is obtained, and then the overall safety status is judged. Design a non-singular distributed fixed-time controller, which is used to achieve fast safety monitoring and intelligent early warning of the tower crane system. The design goal of the controller is to quickly converge the state of the tower crane system to the target safety state. Assume that the current state of the tower crane is , the target safety state is , the state deviation is , and the input of the controller is:

[0055] where, is the inverse matrix of the dynamic function of the tower crane , , , are the gain matrices of the controller, (0, 1) and are design parameters. The gain matrix of the controller is designed as a diagonal matrix, and the specific gain matrix design is as follows:

[0056]

[0057]

[0058] where , , , and are the largest eigenvalue and the second smallest eigenvalue of the Laplacian matrix respectively. These gain matrices are designed through the eigenvalues of the Laplacian matrix to ensure the coordination between tower cranes and the stability of the system. To ensure that the system can converge within a fixed time, and are selected as design parameters to ensure the convergence speed and accuracy of the tower crane state within a finite time. According to the risk metric function and the nonsingular distributed fixed-time controller, a hierarchical early warning mechanism is designed. The risk level is divided into four levels: safe (green), attention (yellow), warning (orange), and danger (red). The risk metric function is divided into four levels according to the set thresholds, and the thresholds are set as:

[0059] When the risk metric function , the tower crane is in the safe level, and the system does not trigger any early warning; when , the tower crane is in the attention level, and the system sends a visual prompt to the operator; when , the tower crane is in the warning level, and the system triggers an audible and visual alarm and restricts the operating speed of the tower crane; when , the tower crane is in the danger level, and the system triggers an emergency brake and locks the movement of the tower crane to prevent accidents.

[0060] In a specific embodiment, the method for global safety monitoring and intelligent early warning of tower crane operations at the construction site further includes the following steps: Establish a three-dimensional space coordinate system according to the actual layout of the construction site and the geometric parameters of the tower crane, and use laser scanning technology to obtain the three-dimensional point cloud data of the buildings and equipment at the construction site to obtain a high-precision 3D map of the construction site; Based on the high-precision 3D map of the construction site and the CAD model, construct a tower crane structure model, a building model, and an obstacle model to obtain a three-dimensional model of the construction site environment; Perform kinematic and dynamic modeling of tower cranes on the three-dimensional model of the construction site environment, including the motion equations and dynamic parameters of the three degrees of freedom of slewing, luffing, and hoisting, to obtain the kinematic and dynamic model of the tower crane; Integrate the global state of the tower crane, the safety risk classification and early warning instructions, and the kinematic and dynamic model of the tower crane into the digital twin platform, and construct a hierarchical architecture including a data layer, a model layer, a service layer, and an application layer to obtain the digital twin model of the tower crane on the construction site; Design a collision detection module and a trajectory planning module based on the digital twin model of the tower crane on the construction site. The collision detection module uses the GJK algorithm to calculate the minimum distance and collision risk between tower cranes in real time. The trajectory planning module uses the A* algorithm and the dynamic window method to consider obstacle constraints and dynamic constraints to obtain the tower crane collision risk assessment and safe trajectory planning scheme; According to the tower crane collision risk assessment and safe trajectory planning scheme, use the multi-agent reinforcement learning algorithm to generate a collaborative risk avoidance strategy to obtain the multi-tower crane collision avoidance collaborative strategy.

[0061] Specifically, establish a three-dimensional space coordinate system according to the actual layout of the construction site and the geometric parameters of the tower crane. The geometric parameters of the tower crane are described by the three-dimensional point cloud data obtained by the laser scanning technology. Assume that the three-dimensional point cloud data of the buildings and equipment on the construction site is , and each point represents the position of a building or equipment. Generate a high-precision 3D map of the construction site through these point cloud data. According to the high-precision 3D map of the construction site and the CAD model, establish the three-dimensional models of the tower crane structure, buildings, and obstacles. Assume that the structural model of the tower crane is , the building model is , and the obstacle model is . These models are used to construct a three-dimensional representation of the construction site environment to help calculate the movement path, collision detection, and trajectory planning of the tower crane. Perform kinematic and dynamic modeling of the tower crane on the three-dimensional model of the construction site environment. The kinematic and dynamic modeling of the tower crane is used to describe the three degrees of freedom of slewing, luffing, and hoisting of the tower crane. Assume that the slewing angle of the tower crane is , and its change rate is The length of the tower crane boom is , and the change rate is The hoisting height of the tower crane is , and the change rate is . Through these motion equations, describe the basic actions of the tower crane. The dynamic parameters of the tower crane include the mass of the tower crane and external forces (such as wind load, load, etc.). Assume that the external force is , then the acceleration of the tower crane is expressed as:

[0062] Among them, is the external force, is the mass of the tower crane, is the acceleration of the tower crane. These parameters and equations help describe the dynamic behavior of the tower crane at the construction site. Integrate the global state, safety risk classification and early warning instructions, and kinematic and dynamic models of the tower crane into the digital twin platform, and monitor the state of the tower crane in real time through this platform and generate real-time decisions. Assume that the global state of the tower crane is represented as , where is the number of tower cranes, represents the state of tower crane . The risk classification and early warning instruction is used to represent the real-time safety state of the tower crane and is calculated by the following formula:

[0063] where, is the risk measure of tower crane , representing the deviation between the current state of the tower crane and its safety domain. Based on the digital twin model of the tower crane, a collision detection module and a trajectory planning module are designed. The GJK algorithm is used to calculate the minimum distance between tower cranes in real time and evaluate the collision risk. Assume that the minimum distance between tower cranes is:

[0064] where, and are the point sets of tower crane and tower crane respectively, represents the minimum distance between them. If is less than the set threshold, it is judged that a collision has occurred. The trajectory planning module uses the A* algorithm and the dynamic window method to consider obstacle constraints and dynamic constraints. The A* algorithm calculates the optimal path from the starting point to the ending point, and its cost function is:

[0065] where, is the actual cost from the starting point to node , is the heuristic estimated cost from node to the target. The dynamic window method is used to adjust the feasible path according to the speed and acceleration limits of the tower crane. According to the tower crane collision risk assessment and safety trajectory planning scheme, a multi-agent reinforcement learning algorithm is used to generate a cooperative risk avoidance strategy. Assume that each tower crane is an agent, its state is , and the action is , then the reward function of the tower crane is represented as:

[0066] Among them, is the distance from the tower crane state to the safe area, c is the collision penalty, and are weight coefficients. Through this reward function, the tower crane learns how to cooperate in a multi-tower crane operation environment, avoid collisions, and ensure safe operation. Through the above steps, the collaborative risk avoidance strategy of the tower crane can ensure that multi-tower cranes avoid collisions and ensure safety during operation, improving the overall operation efficiency.

[0067] The above describes the global safety monitoring and intelligent early warning method for tower crane operations in the embodiments of the present application. Next, the global safety monitoring and intelligent early warning system 10 for tower crane operations in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the global safety monitoring and intelligent early warning system 10 for tower crane operations in the embodiments of the present application includes: A deployment module 11, configured to deploy a multi-level sensor network at multiple target parts of the tower crane at the construction site and collect tower crane operation state data; A filtering processing module 12, configured to perform extended Kalman filtering processing on the tower crane operation state data to obtain tower crane sensor installation angle parameters and non-integrity constraint boom parameters; A state estimation module 13, configured to perform global state estimation based on the tower crane sensor installation angle parameters and non-integrity constraint boom parameters to obtain the tower crane global state and external disturbance data; A safety boundary calculation module 14, configured to input the tower crane global state and external disturbance data into a deep neural network to perform safety boundary calculation to obtain a dynamic safety boundary model of the tower crane at the construction site; A risk metric calculation module 15, configured to perform risk metric calculation based on the dynamic safety boundary model to obtain a safety risk classification early warning instruction.

[0068] Through the collaborative cooperation of the above-mentioned various components, the all-round collection of the operating state data of tower cranes is realized through a multi-level sensor network, providing a high-quality data basis for state estimation; the problem of sensor installation error is solved based on the non-integrity constraint model and extended Kalman filtering technology, improving the accuracy of motion trajectory prediction; the distributed fixed-time observer enables each tower crane to quickly estimate the spatial state parameters and external disturbances under the condition of being unable to obtain the global state, and the upper bound of the early warning response time is not affected by the initial state; the deep learning dynamic safety boundary model breaks through the limitation of traditional fixed-threshold monitoring and realizes the adaptive generation of the safety domain; the combination of the non-singular distributed fixed-time controller and the risk metric calculation method realizes accurate risk quantification and hierarchical early warning; the integrated application of digital twin technology and global collaborative processing, through high-precision three-dimensional modeling and multi-agent reinforcement learning algorithms, generates the optimal multi-tower crane collision avoidance and cooperation strategy, significantly improving the safety and cooperation of tower crane operations at the construction site.

[0069] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0070] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0071] As described above, the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A global safety monitoring and intelligent early warning method for tower crane operations at a construction site, characterized in that: The method comprises: Deploy a multi-layer sensor network at multiple target locations of tower cranes on construction sites and collect data on the operating status of the tower cranes; Performing extended Kalman filter processing on the tower crane operation status data to obtain tower crane sensor installation angle parameters and non-integrity constraint arm parameters; Performing global state estimation based on the tower crane sensor installation angle parameters and the non-integrity constraint arm parameters to obtain the tower crane global state and external disturbance data; Inputting the global state of the tower crane and the external disturbance data into a deep neural network, performing safety boundary calculation, and obtaining a dynamic safety boundary model of the construction site tower crane; A risk measurement calculation is performed based on the dynamic security boundary model to obtain a security risk classification warning instruction.

2. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The multi-level sensor network is deployed at multiple target locations of the tower crane on the construction site, and the tower crane operation status data is collected, including: Install strain sensors on the tower crane boom of the construction site, install angle sensors and MEMS sensors on the slewing mechanism, install triaxial acceleration sensors on the boom end and counterweight end, install weather stations at the four corners of the construction site, and install load sensors in the driver's cab to obtain a multi-level sensor network structure; Performing parameter configuration on the strain sensor, the angle sensor, the acceleration sensor, the weather station and the load sensor, and collecting original operation status data based on the multi-level sensor network; The original operation status data is transmitted through an industrial-grade wireless transmission module to obtain a first operation status data stream, and the first operation status data stream is subjected to wavelet transform denoising and Kalman filtering to obtain a second operation status data stream; The second operation status data stream is compressed to obtain compressed data, and the compressed data is stored in a distributed database according to a unified timestamp and data format to obtain tower crane operation status data.

3. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The extended Kalman filter processing is performed on the tower crane operation status data to obtain the tower crane sensor installation angle parameters and the non-integrity constraint arm parameters, including: Extracting configuration vectors from the tower crane operation status data and establishing a non-integrity constraint basic model for the tower crane; Calculating a sensor installation error matrix according to the tower crane non-integrity constraint basic model, and constructing a state estimator based on the sensor installation error matrix and the three-dimensional position data of the tower crane boom; Constructing a prediction equation and an observation equation for the state estimator to obtain an extended Kalman filter parameter model; Implementing an adaptive adjustment strategy based on the extended Kalman filter parameter model, updating the process noise covariance and the observation noise covariance, and obtaining an adaptive extended Kalman filter; The adaptive extended Kalman filter is iteratively calculated, and the iteration is stopped when the estimated parameter change rate is less than a preset threshold or the number of iterations reaches M times, so as to obtain the tower crane sensor installation angle parameters and the non-integrity constraint arm parameters.

4. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The global state estimation is performed based on the tower crane sensor installation angle parameter and the non-integrity constraint arm parameter to obtain the tower crane global state and external disturbance data, including: The multi-site tower cranes on the construction site are modeled as a network system with N nodes, and a tower crane network communication model is constructed; Based on the tower crane network communication model and the tower crane sensor installation angle parameters, a state equation of the construction site tower crane is constructed to obtain a tower crane dynamic system model; Designing a distributed fixed-time observer structure according to the tower crane dynamic system model and the non-holonomic constraint arm parameters to obtain a distributed fixed-time observer; Designing a gain matrix of the distributed fixed-time observer to obtain an observer gain matrix; Based on the observer gain matrix, the finite time dynamic scale disturbance observer is designed by adopting the immersion and invariance theory to obtain the disturbance observer; According to the distributed fixed-time observer and the disturbance observer, the upper bound of the estimated time is calculated, and iterative calculation is performed within the fixed time to obtain the global state of the tower crane and the external disturbance data.

5. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The global state of the tower crane and the external disturbance data are input into a deep neural network to perform safety boundary calculation to obtain a dynamic safety boundary model of the construction site tower crane, including: Determine a key parameter set for safe operation of the tower crane based on the global state of the tower crane and the external disturbance data, and define a safety domain for each parameter in the key parameter set to obtain a multi-dimensional safety domain; A deep neural network structure is constructed based on the multi-dimensional safety domain. The input layer of the deep neural network structure includes e neurons corresponding to the crane state and external disturbance parameters, the hidden layer includes 4 layers, and the number of neurons in each layer is 256, 128, 64, and 32 respectively. The output layer includes 2m neurons corresponding to the upper and lower limits of the safety domain of m parameters; Performing data enhancement processing based on the global state of the tower crane and the external disturbance data to obtain a neural network training data set, and designing a loss function for the deep neural network structure to obtain a neural network training objective function; Inputting the neural network training data set and the neural network training objective function into the Adam optimizer to execute the cosine annealing learning rate strategy to obtain a trained deep neural network model; The global state of the tower crane and the external disturbance data are input into the trained deep neural network model, and the upper and lower limits of the safety domain of m parameters are output to form an adaptive safety boundary, thereby obtaining a dynamic safety boundary model of the construction site tower crane.

6. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The performing risk measurement calculation based on the dynamic security boundary model to obtain a security risk classification warning instruction includes: Designing a single-machine risk measurement function for a construction site tower crane based on the dynamic safety boundary model, and constructing a global risk measurement function for multiple construction site tower cranes based on the single-machine risk measurement function; Designing a non-singular distributed fixed-time controller based on the dynamic safety boundary model, and designing a gain matrix of the non-singular distributed fixed-time controller to obtain a controller gain matrix; Dividing the risk level into four levels according to the global risk measurement function of tower cranes at multiple construction sites to obtain a risk grading standard; According to the risk classification standard and the non-singular distributed fixed time controller, early warning measures are formulated for different risk levels. The safety level does not trigger an early warning, the attention level sends a visual prompt to the operator, the warning level triggers an audible and visual alarm and limits the tower crane operating speed, and the danger level triggers an emergency brake and locks the tower crane movement, thereby obtaining a safety risk classification early warning instruction.

7. The global safety monitoring and intelligent early warning method for tower crane operation at a construction site according to claim 1 is characterized in that: The global safety monitoring and intelligent early warning method for tower crane operation at a construction site also includes: A three-dimensional spatial coordinate system is established based on the actual layout of the construction site and the geometric parameters of the tower crane. The three-dimensional point cloud data of the construction site buildings and equipment are obtained using laser scanning technology to obtain a high-precision 3D map of the construction site. Based on the high-precision 3D map and CAD model of the construction site, a tower crane structure model, a building model and an obstacle model are constructed to obtain a three-dimensional model of the construction site environment; Performing kinematic and dynamic modeling of the tower crane on the three-dimensional model of the construction site environment, including motion equations and dynamic parameters of three degrees of freedom: rotation, amplitude change, and lifting, to obtain a kinematic and dynamic model of the tower crane; Integrate the global state of the tower crane, the safety risk classification warning instructions and the kinematics and dynamics model of the tower crane into the digital twin platform, build a layered architecture including a data layer, a model layer, a service layer and an application layer, and obtain a digital twin model of the tower crane on the construction site; Based on the digital twin model of the tower crane on the construction site, a collision detection module and a trajectory planning module are designed. The collision detection module uses the GJK algorithm to calculate the minimum distance and collision risk between tower cranes in real time. The trajectory planning module uses the A* algorithm and the dynamic window method to consider obstacle constraints and dynamic constraints to obtain a tower crane collision risk assessment and a safe trajectory planning scheme. According to the tower crane collision risk assessment and the safety trajectory planning scheme, a multi-agent reinforcement learning algorithm is used to generate a collaborative risk avoidance strategy to obtain a multi-tower crane collision avoidance collaborative strategy.

8. A global safety monitoring and intelligent early warning system for tower crane operations at construction sites, characterized in that: A method for global safety monitoring and intelligent early warning of a construction site tower crane operation according to any one of claims 1 to 7, wherein the global safety monitoring and intelligent early warning system for a construction site tower crane operation comprises: A deployment module is used to deploy a multi-level sensor network at multiple target locations of tower cranes on construction sites and collect data on the operation status of the tower cranes; A filtering processing module is used to perform extended Kalman filtering on the tower crane operation status data to obtain tower crane sensor installation angle parameters and non-integrity constraint arm parameters; A state estimation module, used for performing global state estimation based on the tower crane sensor installation angle parameters and the non-integrity constraint arm parameters to obtain the tower crane global state and external disturbance data; A safety boundary calculation module is used to input the global state of the tower crane and the external disturbance data into a deep neural network to perform safety boundary calculation to obtain a dynamic safety boundary model of the construction site tower crane; The risk measurement calculation module is used to perform risk measurement calculation based on the dynamic security boundary model to obtain a security risk classification warning instruction.

Images