Real-time early warning method and system for double-crane lifting

Abstract

The invention relates to the technical field of information, and discloses a real-time early warning method and system for double-crane hoisting, and the method comprises the steps: obtaining the operation condition data of a plurality of hoisting devices in real time, and carrying out the standardized packaging, and obtaining heterogeneous processing data; on the basis of the heterogeneous processing data, constructing a scene simulation model through a digital twinning technology so as to output collaborative parameters of the hoisting equipment in various business scenes; inputting the collaborative parameters and the operation condition data into a pre-constructed double-layer early warning model for processing, and outputting an early warning result; wherein the double-layer early warning model comprises a hybrid anomaly detection model integrating an isolated forest and a single-class support vector machine and an LSTM network model introducing an attention mechanism; and real-time and accurate early warning of double-crane lifting is realized through a digital twinborn technology and a machine learning algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, in particular to a real-time early warning method and system for double-machine lifting. BACKGROUND

[0002] In modern engineering construction, double-machine lifting has become an indispensable key link with its strong carrying capacity. However, the cooperation of two giants also means double risks and challenges. The traditional operation mode mainly relies on the personal experience and visual observation of the operator to judge the operation state, which is difficult to realize real-time and comprehensive perception of potential dangerous factors. At the same time, in the prior art, the data of each crane is usually isolated in the scene of double-machine cooperative lifting, lacking real-time linkage and comparative analysis capability of the working condition data of each device, and unable to realize real-time warning according to the specific needs of cooperative operation. SUMMARY

[0003] The present application provides a real-time early warning method and system for double-machine lifting, which solves the problem of how to realize real-time and accurate early warning for double-machine lifting.

[0004] To solve the above technical problems, the present application provides a real-time early warning method for double-machine lifting, comprising:

[0005] Real-time acquisition of running condition data of multiple hoisting devices and standardized packaging to obtain heterogeneous processing data;

[0006] Based on the heterogeneous processing data, a scene simulation model is constructed by digital twinning technology to output the cooperative parameters of each hoisting device in multiple business scenarios;

[0007] Each of the cooperative parameters and the running condition data is input into a pre-constructed double-layer early warning model for processing to output an early warning result; wherein the double-layer early warning model includes a hybrid anomaly detection model integrating isolated forest and one-class support vector machine, and an LSTM network model introducing attention mechanism.

[0008] As one of the preferred schemes, the real-time acquisition of running condition data of multiple hoisting devices and the standardized packaging to obtain heterogeneous processing data, comprising:

[0009] Real-time acquisition of running condition data of multiple hoisting devices, and determination of the operation amplitude of each hoisting device according to each running condition data;

[0010] Calling a device performance library based on each running condition data to real-time match the rated lifting weight of each hoisting device in the current posture;

[0011] The operation condition data, the operation range and the rated lifting weight are converted into a unified JSON format, and a timestamp is attached to obtain heterogeneous processing data.

[0012] As one of the preferred solutions, the scene simulation model is constructed by digital twin technology based on the heterogeneous processing data, including:

[0013] Based on the heterogeneous processing data, each business scenario is deconstructed to identify the physical entities of each business scenario, and the data sources of each physical entity are determined to construct a mapping table between the physical entities and data sources;

[0014] According to the mapping table, three-dimensional geometric modeling is performed on each hoisting device, and kinematic skeleton binding is performed on the modeling results to generate a kinematic model of each hoisting device;

[0015] A virtual-real synchronization mechanism and a phase compensation coefficient are introduced to map the kinematic model and the heterogeneous processing data to obtain a twin body of each hoisting device and perform multi-body collaborative aggregation to generate a twin aggregate model;

[0016] The test scene is simulated by the twin aggregate model to quantify the error between the simulation results and the actual results, and the phase compensation coefficient is adjusted when the error does not meet the preset error;

[0017] The construction steps of the twin aggregate model are iteratively executed according to the adjustment results until the error meets the preset error, and the final twin aggregate model is taken as the scene simulation model.

[0018] As one of the preferred solutions, according to the mapping table, three-dimensional geometric modeling is performed on each hoisting device, and kinematic skeleton binding is performed on the modeling results to generate a kinematic model of each hoisting device, including:

[0019] A three-dimensional geometric model of each hoisting device is constructed by a three-dimensional modeling algorithm, and key structures are reproduced and environmental modeling is performed to obtain a three-dimensional spatial structure model of each hoisting device;

[0020] A kinematic skeleton is constructed inside each three-dimensional spatial structure model to define the connection relationship and operation freedom degree between the components of each hoisting device to generate an operation connection model of each hoisting device;

[0021] Based on the mapping table, the heterogeneous processing data is mapped to the skeleton parameters of each operation connection model, and physical properties are assigned to obtain a kinematic model of each hoisting device.

[0022] As one of the preferred solutions, the multi-body cooperative polymerization is performed to generate a twin polymer model, which includes:

[0023] Each of the twins is loaded into the same business scenario, and each of the twins is aligned in absolute coordinates;

[0024] According to the business logic of the business scenario, a multi-point constraint relationship between the hoisted object and each of the hoisting equipment is established;

[0025] Through the physical engine and the script corresponding to the business logic, the cooperative behavior of each of the hoisting equipment under the multi-point constraint relationship is simulated to generate the twin polymer model.

[0026] As one of the preferred solutions, the business scenario includes a common lifting scene, and the cooperative parameters include a single machine load distribution rate and a single machine load; wherein,

[0027] The output of the cooperative parameters of each of the hoisting equipment in multiple business scenarios includes:

[0028] After the heterogeneous processing data is time-space synchronized and calibrated, it is input to the scene simulation model processing to align the spatial coordinates of the hoisted object and each of the hoisting equipment in the common lifting scene, and to output the single machine load distribution rate and the single machine load by combining the influence of the physical engine simulation single machine action on the distribution rate.

[0029] As one of the preferred solutions, the business scenario also includes a master-slave lifting scene, and the cooperative parameters include a work stage and a real-time pitch angle of the hoisted object; wherein,

[0030] The output of the cooperative parameters of each of the hoisting equipment in multiple business scenarios includes:

[0031] After the heterogeneous processing data is time-space synchronized and calibrated, it is input to the scene simulation model processing to associate each of the hoisting equipment with the inclination angle of the hoisted object based on the business logic of the master-slave lifting scene through kinematic coupling, to obtain the real-time pitch angle of the hoisted object and compare it with a preset inclination threshold to determine the work stage.

[0032] As one of the preferred solutions, the input of each of the cooperative parameters and the operating condition data into the pre-constructed double-layer early warning model for processing to output a warning result includes:

[0033] A first cooperative work dynamic feature is determined according to the single machine load distribution rate and the single machine load, and a sliding window statistical feature is constructed according to the operating condition data;

[0034] The first cooperative operation dynamic feature and the sliding window statistical feature are input into the LSTM network model with the introduced attention mechanism for processing, and a single-machine load prediction distribution rate and a single-machine predicted load of each hoisting device are output.

[0035] When the single-machine load prediction distribution rate exceeds a preset distribution rate threshold, a deviation between the single-machine load prediction distribution rate and the preset distribution rate threshold is calculated, and load differences between the single-machine predicted loads are quantified.

[0036] A warning result is generated according to the deviation and the load differences for output.

[0037] As one of the preferred solutions, the cooperative parameters and the operation condition data are input into a pre-constructed double-layer warning model for processing, and a warning result is output, which includes:

[0038] A second cooperative operation dynamic feature is determined according to the real-time pitch angle, and the operation condition data are input into the hybrid anomaly detection model for processing, so as to obtain a first anomaly score corresponding to the isolated forest and a second anomaly score corresponding to the single-class support vector machine.

[0039] According to the operation stage, a first weight and a second weight are determined in a preset weight library, so as to weight and fuse the first anomaly score and the second anomaly score, and obtain a final anomaly score.

[0040] The final anomaly score is compared with a preset warning threshold, and a warning result is generated according to a comparison result for output.

[0041] The second aspect of the present application provides a real-time warning system for double-machine lifting, which comprises:

[0042] A data processing module is configured to acquire operation condition data of a plurality of hoisting devices in real time and perform standardized packaging to obtain heterogeneous processing data.

[0043] A model simulation module is configured to construct a scene simulation model based on the heterogeneous processing data by using a digital twinning technology, so as to output cooperative parameters of each hoisting device in a plurality of business scenarios.

[0044] A result output module is configured to input the cooperative parameters and the operation condition data into a pre-constructed double-layer warning model for processing, and output a warning result, wherein the double-layer warning model comprises a hybrid anomaly detection model integrating an isolated forest and a single-class support vector machine, and an LSTM network model with an introduced attention mechanism.

[0045] Compared with the prior art, the embodiment of the present application has the following advantages:

[0046] By acquiring the running working condition data of the hoisting equipment in real time, the scene simulation model is dynamically updated in combination with the digital twinning technology, so that the collaborative parameters are highly matched with the actual working conditions, and the collaborative failure caused by data lag is reduced; the fusion of heterogeneous data solves the problem of non-uniform data format, improves the model input quality, and provides a reliable basis for subsequent anomaly detection; unknown anomalies are quickly identified by using a hybrid anomaly detection model (isolated forest + single-class SVM), and the time sequence dependence is captured by introducing an attention mechanism LSTM network, so that real-time and accurate early warning of double-machine lifting is realized by the digital twinning technology and the machine learning algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings described in the following are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0048] Figure 1 is a flow chart of a real-time early warning method for double-machine lifting provided by an embodiment of the present application;

[0049] Figure 2 is a structural diagram of a real-time early warning system for double-machine lifting provided by an embodiment of the present application;

[0050] Reference signs:

[0051] Among them, 10, data processing module; 20, model simulation module; 30, result output module. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings and embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. The purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0053] In the description of the present application, the terms "first", "second", "third" and the like are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0054] In the description of the present application, it should be noted that unless otherwise explicitly defined and limited, the terms "mounting", "connection", "connecting" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the system or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be construed as limiting the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0055] In the description of the present application, it should be noted that unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as generally understood by those skilled in the art. The terms used in the specification of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0056] In an embodiment, as shown in Figure 1 The first aspect of the present application provides a real-time early warning method for double-machine lifting, comprising:

[0057] S1, real-time acquisition of operation condition data of a plurality of hoisting devices and standardized packaging, to obtain heterogeneous processing data;

[0058] In an embodiment, step S1 comprises:

[0059] Real-time acquisition of operation condition data of a plurality of hoisting devices, and determination of the working amplitude of each hoisting device according to each operation condition data;

[0060] Calling a device performance library based on each operation condition data to real-time match the rated hoisting weight of each hoisting device in the current posture;

[0061] Converting each operation condition data, each working amplitude and each rated hoisting weight into a unified JSON format, and adding a time stamp to obtain heterogeneous processing data.

[0062] Specifically, through the collection terminal composed of various sensors such as gateways, weight / tilt angle / GPS / pose / wind speed sensors, wireless transceiver terminals, sound and light alarms and the like, the running condition data of the hoisting equipment is collected in real time, such as the real-time hoisting weight, real-time tilt angle, real-time coordinates, real-time rotation angle and the like of each hoisting equipment, and the environmental data of the operation site is also collected, such as wind speed, wind direction, obstacle coordinates of the work site (determined by laser radar), temperature and humidity (auxiliary parameters affecting equipment performance) and the like, and the static data of the hoisting equipment is obtained, such as the machine model, equipment ID, boom length, rated hoisting weight and the like, and the basic parameters of the hoisted object, such as the hoisted object weight, hoisted object size, hoisted object centroid coordinates and the like; after the sensor collects the data, the 4G communication module integrated through the gateway is used to upload the data to the central processing platform in real time, so that the central processing platform determines the early warning result through processing of the data, and then the sound and light alarm is used for alarm; wherein the hoisting equipment includes cranes, elevators and the like, and has two or more, and the present application takes two cranes as an example to illustrate the real-time early warning method for double-machine lifting; then, based on the running condition data of each hoisting equipment, the work range is obtained through work range = arm length x cos(rotation angle) + basic offset (i.e. the distance from the center of the crane to the rotation center), and for a crawler crane: work range = boom projection length x cos(boom angle) + outrigger span correction value, and then the work range of each hoisting equipment is obtained.

[0063] The equipment performance library is stored in the form of a database (such as MySQL) or a configuration file, and contains: equipment model, arm length segmentation, rotation angle range, wind speed limit, rated hoisting weight table (such as the rated hoisting weight of QY25V crawler crane is 25 tons at 10-meter range and 0° tilt angle), and environmental correction coefficient (such as hoisting weight reduction by 10% in high-temperature weather), and the current equipment ID, work range, boom tilt angle, wind speed and environmental temperature of the hoisting equipment are taken as inputs, the performance table of the corresponding model is selected according to the equipment ID, the record meeting the conditions of "work range ≤ maximum allowed range" and "tilt angle within the effective range" is found, and the environmental correction coefficient (such as rated hoisting weight x 0.9 when the temperature is greater than 35°C) is applied, to obtain the real-time rated hoisting weight of each hoisting equipment under the current posture.

[0064] The running condition data, work range and rated hoisting weight of each hoisting equipment are integrated into a unified JSON format, and a timestamp is added to obtain heterogeneous processing data.

[0065] The present application matches the rated hoisting weight under the current posture in real time through the equipment performance library, avoids overloading operation and improves safety; calculates the work range based on the running condition data to provide key geometric parameters for collaborative operation and reduce the risk of collision; converts multi-source heterogeneous data into JSON format to solve the problem of non-uniform data format and facilitate subsequent processing; and adds a timestamp to make the data associated with a specific operation time, thereby supporting accident backtracking and trajectory analysis.

[0066] S2, constructing a scene simulation model based on the heterogeneous processing data by a digital twin technology to output coordination parameters of each hoisting device in multiple business scenarios;

[0067] In an embodiment, the constructing a scene simulation model based on the heterogeneous processing data by a digital twin technology comprises:

[0068] deconstructing each business scenario based on the heterogeneous processing data to identify physical entities of each business scenario and determine data sources of each physical entity to construct a mapping table between the physical entities and the data sources;

[0069] performing three-dimensional geometric modeling on each hoisting device according to the mapping table and kinematic skeleton binding on the modeling result to generate a kinematic model of each hoisting device;

[0070] introducing a virtual-real synchronization mechanism and a phase compensation coefficient to perform parameter mapping on each kinematic model and the heterogeneous processing data, obtain a twin body of each hoisting device, and perform multi-body coordination aggregation to generate a twin aggregated body model;

[0071] simulating a test scene through the twin aggregated body model to quantify an error between a simulation result and an actual result, and adjusting the phase compensation coefficient when the error does not meet a preset error;

[0072] iteratively performing the construction steps of the twin aggregated body model according to the adjustment result until the obtained error meets the preset error, and taking the finally obtained twin aggregated body model as the scene simulation model.

[0073] Specifically, based on the heterogeneous processing data, the application adopts a function-structure-component (FSC) hierarchical model 3D modeling tool or a property extraction method in SolidWorks, uses a natural language processing technology (NLP) to deconstruct construction texts corresponding to each business scenario, extracts key physical entities, which include a crane entity: identifies its model, static geometric parameters (arm length range, tower height, etc.), dynamic parameters (such as performance tables provided by manufacturers), and variables that need to be monitored in real time (such as boom luffing angle, slewing angle, hook height, etc.); a hoisted object entity: determines its weight, geometric size, center of mass position, and hoisting point distribution; an environment entity: identifies environmental factors that affect work safety, such as wind speed, fixed obstacles in the work area, etc.; then determines the data sources corresponding to each physical property that needs to be monitored, i.e., specific sensor types (for example: luffing angle → inclination sensor, hoisting weight → weight sensor, wind speed → anemometer), to establish a mapping table between physical entities and data sources.

[0074] Using industrial-grade three-dimensional modeling software (such as AutoCAD, SolidWorks, CATIA, etc.) or game engine-based development platforms (such as Unity, Unreal Engine), create high-fidelity three-dimensional models for each crane and typical hoisted object, accurately reflecting their actual size, structure, and appearance, and define the kinematic skeleton (URDF file) in ROS, binding the three-dimensional model to joints (such as slewing joints, luffing cylinders), publishing joint states through the tf2 library of ROS, achieving kinematic control of the model, and thus obtaining a device model that supports kinematic simulation, i.e., a kinematic model.

[0075] Define a standard JSON or similar format data model to convert heterogeneous sensor data uploaded from edge gateways into unified, structured data that can be understood by the twin platform, and then based on virtual-real synchronization mechanisms, accurately map standardized real-time data streams to corresponding parameters in the twin model through the MQTT protocol, such as binding the data stream of the inclination sensor to the pitch angle parameter of the crane twin's boom model, wind speed to environmental wind field strength, and hoisted weight data to model load force, etc. Then, using the rendering engine of the twin platform, update the position, pose, and state of each component of the model in real time at a high refresh rate of 50Hz based on the continuously flowing data, achieving synchronization between the virtual world and the physical world. Then, to compensate for network latency, introduce a phase compensation coefficient, whose formula is: model update time = physical collection time + transmission delay x compensation coefficient, to offset data transmission lag, and thus obtain the twin of each hoisting device, and aggregate multiple device models (such as through Unity's GameObject hierarchical management) to generate a twin aggregate model.

[0076] Simulate typical working conditions (such as hoisting a 5-ton load to a 10-meter amplitude) in Unity, record the output data of the twin model (such as boom stress, amplitude), and then compare it with the actual sensor data to calculate the error (the difference between the simulation result and the actual result compared with the actual result). If the error is greater than the preset threshold (such as 1%), return to the previous step to calibrate the model, for example, use lead-lag compensator, adaptive phase compensation (based on error gradient), or fuzzy control algorithm to correct the compensation coefficient, adjust the physical engine parameters, etc., and repeat the construction steps of the twin aggregate model based on the calibrated data until the model's behavior is highly consistent with the real world and meets the preset error, and the final twin aggregate model is used as the scene simulation model.

[0077] The application ensures high consistency of the digital twin model and the actual equipment by deconstructing the business scene and establishing a mapping table of physical entities and data sources; introduces a virtual-real synchronization mechanism and a phase compensation coefficient, dynamically corrects the model parameters, and significantly reduces the simulation error; performs three-dimensional geometric modeling and kinematic skeleton binding on the hoisting equipment, supports multi-device collaborative operation simulation, avoids the high cost and risk of physical experiments; through quantitative simulation error and dynamic adjustment of the phase compensation coefficient, the closed-loop optimization of the model is realized, and finally the scene simulation model conforming to the actual working condition is generated; based on the heterogeneous data processing and modular modeling method, different types of cranes or complex construction environments can be quickly adapted, and the scalability is strong.

[0078] In an embodiment, the three-dimensional geometric modeling of each hoisting equipment according to the mapping table is performed, and the kinematic skeleton binding of the modeling result is performed to generate a kinematic model of each hoisting equipment, comprising:

[0079] The three-dimensional geometric model of each hoisting equipment is constructed by a three-dimensional modeling algorithm, and the key structure is reproduced and the environment modeling is performed to obtain a three-dimensional space structure model of each hoisting equipment;

[0080] The kinematic skeleton is constructed inside each three-dimensional space structure model to define the connection relationship and the running freedom degree between the components contained in each hoisting equipment, and a running connection model of each hoisting equipment is generated;

[0081] Based on the mapping table, the heterogeneous processing data is mapped to the skeleton parameters of each running connection model, and the physical properties are given to obtain a kinematic model of each hoisting equipment.

[0082] Specifically, the application can also use entity modeling, parameter modeling, AutoCAD (accurate size) + 3ds Max / Blender (detail optimization) and other three-dimensional modeling algorithms to construct a high-fidelity model. For cranes: restore the cross section of the boom, the hinge point structure, and the shape of the hook; for the hoisted object: model according to the actual shape (such as box structure, cylindrical structure, etc.), and mark the center of mass coordinates, then simplify the surface number + LOD (level of detail) to optimize the model for lightweight processing to ensure real-time rendering performance; then mark the hinge points of the boom and the slewing ring, the connection points of the hook and the hoisted object, etc. to reproduce the key structure, and then import the job site CAD drawing to generate the three-dimensional boundaries of the ground and obstacles (such as factory column, overhead cable), to perform environment modeling, and thus obtain the three-dimensional space structure model of each hoisting equipment.

[0083] A kinematic skeleton is constructed inside the three-dimensional space structure model of each hoisting equipment to define the connection relationship and running freedom (such as the pitch of the boom, the horizontal rotation of the platform) between the components (such as the boom, the hook, the rotating platform) contained in each hoisting equipment, such as creating a kinematic chain of "boom → hinge point → rotating platform" for the crane, creating a constraint chain of "center of mass → lifting point" for the hoisted object, and the like, and further generating a running connection model of each hoisting equipment.

[0084] Based on the constructed mapping table, the heterogeneous processing data is mapped to the skeleton parameters of the running connection model of each hoisting equipment (such as the inclination sensor data → the pitch angle of the boom skeleton, the GPS data → the rotation angle of the rotating platform skeleton, and the like), and the physical properties such as mass, inertia, and collision body are assigned, and a physical engine is integrated to simulate physical effects such as gravity, wind force, inertia, and collision, so as to ensure that the behavior of the virtual model conforms to the physical laws of the real world, and further obtain a kinematic model of each hoisting equipment.

[0085] The present application accurately reproduces the key structures and environment of the hoisting equipment through a three-dimensional modeling algorithm, ensures the consistency of the model and the actual equipment, constructs a kinematic skeleton inside the three-dimensional model, clearly defines the connection relationship and freedom between components, provides a basis for subsequent dynamic simulation, can quickly adapt to different types of equipment through parameterized skeleton binding, realizes dynamic assignment of physical properties based on a mapping table, maps heterogeneous data to skeleton parameters, improves the simulation reality, and is suitable for crane design verification, operation training, and dangerous working condition pre-rehearsal.

[0086] In an embodiment, the performing multi-body collaborative aggregation to generate a twin aggregate model comprises:

[0087] loading each of the twins into the same business scenario, and performing absolute coordinate alignment on each of the twins;

[0088] establishing a multi-point constraint relationship between the hoisted object and each of the hoisting equipment according to the business logic of the business scenario;

[0089] simulating the collaborative behavior of each of the hoisting equipment under the multi-point constraint relationship through a physical engine and a script corresponding to the business logic, to generate the twin aggregate model.

[0090] Specifically, the present application loads independent crane twins, lifting object twins and environment twins into the same virtual scene, and aligns the absolute coordinates of each twin based on GPS positioning to establish spatial coordinate association. Then, logical and physical connections between models are established, and multi-point constraint relationships between lifting objects and each lifting equipment are established according to the business logic of the business scene (i.e. the operation steps corresponding to the business scene, which can be determined and digitized by manual experience). For example, the fixed connection of the hook and the lifting object, the elastic connection of the steel wire rope (which cannot exceed the elastic threshold), the prohibition of amplitude change when the lifting object height is greater than 10m, and other lifting rules. Then, based on the script corresponding to the business logic, the collaborative behavior between each lifting equipment under the multi-point constraint relationship is simulated through the physical engine Gazebo, i.e. how the action of one crane transmits force and influences another through the lifting object (such as lifting the boom of No. 1 crane → tilting the lifting object → sudden increase of load of No. 2 crane), and enabling multi-body collision detection in Gazebo (such as contact force calculation of lifting arm and environmental obstacles), setting time step to ensure stability; recording the state (position, velocity, force) of each twin during simulation, generating time series data set, and packaging dynamic constraint relationship and collaborative logic together with each twin as a standardized twin aggregate model.

[0091] The present application accurately simulates the complex interaction between lifting equipment and lifting objects through absolute coordinate alignment and multi-constraint relationship modeling, avoiding the isolated errors of traditional single-body simulation; defines the constraint relationship according to the actual business scene, so that the simulation result is directly mapped to the real operation specification; uses a physical engine to calculate multi-body dynamics in real time, combined with script logic, to generate collaborative motion trajectories that conform to physical laws; modular design supports the addition of new equipment or adjustment of business logic, and outputs standardized twin aggregate models for easy integration into a digital twin platform.

[0092] In an embodiment, the business scene includes a common lifting scene, and the collaborative parameters include single-machine load distribution rate and single-machine load; wherein,

[0093] The output of the collaborative parameters of each lifting equipment in multiple business scenes includes:

[0094] After the heterogeneous processing data is time and space synchronized and calibrated, it is input to the scene simulation model for processing to align the spatial coordinates of the lifting object and each lifting equipment in the common lifting scene, and to output the single-machine load distribution rate and the single-machine load by combining the physical engine to simulate the influence of single-machine action on the distribution rate.

[0095] Specifically, the application controls the time error of different sensors within 10 ms by isomerization processing data, ensures that the data can be associated in the time dimension, and converts the position data of all devices and hoisted objects into a unified coordinate system (such as a three-dimensional coordinate with the center point of the site as the origin) based on GPS coordinates and site CAD maps, solves the problem of inconsistent position reference of each device, and realizes time-space synchronous calibration to input into the scene simulation model for processing, so that the digital twin model aligns the spatial coordinates of hoisted objects and each hoisting device in the common lifting scene, and quantifies the actual single machine load distribution rate and actual single machine load based on real-time hoisting data, A machine distribution rate = A machine hoisting weight / (A machine hoisting weight and B machine hoisting weight) × 100%, and B machine distribution rate is the same, and the single machine load of each hoisting device under the hoisting weight is measured; then the influence of single machine action on the distribution rate is simulated by using a physical engine to calculate the spatial lever relationship between the centroid of the hoisted object and the double hooks to derive the theoretical distribution rate: theoretical A machine distribution rate = distance from the centroid to the B hook / distance between the double hooks × 100%, and the theoretical B machine distribution rate can be calculated in the same way; the actual single machine load distribution rate and the theoretical single machine load distribution rate are compared, if the deviation is greater than 8%, it is judged that the stress is abnormal, then the construction process of the scene simulation model is repeated until the deviation between the actual single machine load distribution rate and the theoretical single machine load distribution rate is not greater than 8%, and the single machine load distribution rate and the single machine load of each hoisting device at this time are output.

[0096] The application accurately outputs the single machine load distribution rate and the single machine load in the common lifting scene, which helps the operator clearly understand the stress condition of each hoisting device to avoid accidents such as device damage and hoisted object falling caused by uneven load distribution or excessive load, and ensures the safety of personnel and equipment on the construction site; the influence of single machine action on the distribution rate is simulated by using a physical engine to simulate the scene simulation model, which can plan a more reasonable lifting scheme in advance, so that each hoisting device can cooperate more efficiently in the cooperative operation process, reduce operation interruption and repeated operation caused by improper coordination, and improve the overall operation efficiency.

[0097] In an embodiment, the business scene further includes a master-slave lifting scene, and the cooperation parameter includes a working stage and a real-time pitch angle of the hoisted object; wherein,

[0098] The output of the cooperation parameter of each hoisting device in multiple business scenes includes:

[0099] After the isomerization processing data is time-space synchronous calibrated, it is input into the scene simulation model for processing, so as to associate each hoisting device and the inclination angle of the hoisted object through kinematic coupling based on the business logic of the master-slave lifting scene, obtain the real-time pitch angle of the hoisted object, and compare it with the preset inclination threshold to determine the working stage.

[0100] Specifically, the application can also keep all data consistent in time and space dimensions through methods such as time stamp alignment and spatial coordinate conversion, so as to be input into a scene simulation model for processing after time and space synchronization calibration, so that the scene simulation model determines the business logic of the master-slave lifting scene according to the manual experience of the operator, and establishes a mathematical relationship model between each hoisting device and the inclination angle of the center of mass of the hoisted object through kinematic coupling: a fixed global coordinate system is established, taking a certain point on the ground as the origin, and the directions of the three coordinate axes are specified, a local coordinate system is established for the main crane with the center of rotation as the origin, a similar local coordinate system is established for the slave crane, and a local coordinate system is established for the hoisted object with its center of mass as the origin; wherein the local coordinate system is used to describe the motion and attitude change of each component; according to the motion parameters (such as lifting height, rotation angle, amplitude angle, etc.) of the master-slave hoisting equipment, the position coordinates of the master-slave hoisting equipment hooks in the global coordinate system are calculated by using geometric relationships and coordinate transformation principles; then, according to the connection relationship between the hoisted object and the hooks of the master-slave hoisting equipment, a mathematical relationship between the position coordinates of the center of gravity of the hoisted object in the global coordinate system and the position coordinates of the hooks of the master-slave hoisting equipment is established; taking the pitch angle of the hoisted object as an example, the mathematical relationship between the pitch angle and the motion parameters of the master-slave hoisting equipment is established by analyzing the influence of the motion of the master-slave hoisting equipment on the attitude of the hoisted object, and the attitude change of the hoisted object can also be described by establishing a multi-degree-of-freedom kinematic equation. For example, the homogeneous coordinate transformation matrix is used to represent the transformation relationship between the master-slave hoisting equipment and the hoisted object in different coordinate systems, and the mathematical expression between the attitude (pitch angle, yaw angle, roll angle) of the hoisted object and the motion parameters of the master-slave hoisting equipment is obtained through matrix operation.

[0101] Then, based on the established mathematical model, the physical engine is used to simulate the motion of the hoisting equipment and the attitude change of the hoisted object, to obtain the real-time pitch angle of the hoisted object and compare it with the preset inclination threshold to determine the operation stage; wherein the operation stage includes flat lifting, turning over and vertical; the inclination angle is 0°-5° for flat lifting, 10°-45° for turning over, and 85°-90° for vertical.

[0102] The application determines the inclination angle of the target hoisted object according to the characteristics of the master-slave lifting scene: dynamic change of double-machine load, unbalanced hoisting operation, and accurately infers the operation stage through the change of the inclination angle, which can timely discover potential safety hazards; the hoisting equipment and the inclination angle of the hoisted object are associated through kinematic coupling, and the operation stage is accurately determined, which helps the operator to reasonably arrange the operation steps and equipment operation according to the characteristics of different operation stages, improves the operation efficiency, and reduces unnecessary operation and waiting time; based on the business logic and the analysis method of kinematic coupling, the actual relationship between each device and the hoisted object in the master-slave lifting process can be more accurately reflected, the output of the coordination parameters is more accurate, and it is helpful for better cooperation between each hoisting equipment to realize the optimization of collaborative operation.

[0103] S3, inputting each of the cooperative parameters and the operating condition data into a pre-constructed double-layer early warning model for processing, and outputting a warning result; wherein the double-layer early warning model comprises a hybrid anomaly detection model integrating an isolated forest and a one-class support vector machine, and an LSTM network model introducing an attention mechanism;

[0104] In an embodiment, step S3 comprises:

[0105] According to the single-machine load distribution rate and the single-machine load, a first cooperative operation dynamic feature is determined, and a sliding window statistical feature is constructed according to the operating condition data;

[0106] The first cooperative operation dynamic feature and the sliding window statistical feature are input into the LSTM network model introducing an attention mechanism for processing, and a single-machine load prediction distribution rate of each hoisting device and a single-machine prediction load thereof are output;

[0107] When the single-machine load prediction distribution rate exceeds a preset distribution rate threshold, a deviation amount between the single-machine load prediction distribution rate and the preset distribution rate threshold is calculated, and a load difference between the single-machine prediction loads is quantified;

[0108] According to the deviation amount and the load difference, a warning result is generated for output.

[0109] Specifically, the present application designs a double-layer early warning model for two operation scenarios. For a common lifting scenario, an LSTM network model introducing an attention mechanism is used to process a scene simulation model, and then a warning result is generated according to the processing result.

[0110] By introducing a cooperative operation dynamic feature to reflect a double-machine linkage state, a single-machine load distribution rate of each hoisting device, a change rate of a single-machine load in a preset period of time are calculated respectively, a load distribution rate change gradient and a load change gradient are obtained as a first cooperative operation dynamic feature, and based on operating condition data, mean, variance, kurtosis, etc. of key parameters (such as total hoisting weight, single-machine hoisting weight, etc.) of each hoisting device in the past N time points (such as 30 seconds) are quantified to capture short-term trends and volatility, and a sliding window statistical feature is constructed.

[0111] The first cooperative operation dynamic feature and the sliding window statistical feature are input into a double-layer early warning model, which is processed by an LSTM network model with attention mechanism. The model automatically learns the time series features and long-term dependencies in the data, and gives different weights to the data at different time steps through the attention mechanism, highlighting important information, and then outputs the single machine load prediction allocation rate and the single machine prediction load of each hoisting equipment. The LSTM network model with attention mechanism is formed by introducing an attention layer (Attention Layer) after the standard LSTM network layer. The traditional LSTM gives equal treatment to all historical time points when processing long sequences, which easily ignores a few key time points. The attention mechanism allows the model to dynamically assign different "attention weights" to different parts of the input sequence when making predictions, focusing on the historical information that has the greatest impact on future results.

[0112] The calculation of attention weights generally follows the following logic:

[0113] Calculate the relevance score e_t between the hidden state h_t at each time step and the final hidden state h_final of the LSTM network through a small feedforward neural network;

[0114] Use the Softmax function to normalize all scores e_t to get the attention weight a_t;

[0115] Weighted sum of the weight a_t and the corresponding hidden state h_t generates a "context vector" C, which contains an important summary of historical information for prediction;

[0116] Finally, combine the context vector C with the final output of the LSTM and send it to a fully connected layer for final parameter value prediction, i.e., single machine load prediction allocation rate and single machine prediction load.

[0117] The LSTM network model with attention mechanism includes: an input layer for receiving three-dimensional input data; a multi-feature fusion layer (1D convolution) for convolution operation on input features to improve feature expression ability; a bidirectional LSTM layer for capturing forward and backward time sequence dependencies and outputting hidden state sequences; an attention mechanism layer for calculating the weight of each time step and generating a context vector; a fully connected layer for mapping the context vector to the prediction target (single machine load allocation rate and load); and an output layer for outputting multi-task prediction results (allocation rate and load of each hoisting equipment).

[0118] When the output single-machine load prediction allocation rate exceeds the preset allocation rate threshold (which is usually lower than the rated lifting capacity of the equipment (such as 80%-90%) to leave a safety margin), the deviation between the single-machine load prediction allocation rate and the preset allocation rate threshold is calculated, and the load difference between the single-machine prediction loads of each hoisting equipment is quantified to evaluate the balance of load allocation.

[0119] According to the size and trend of the deviation and the load difference, corresponding early warning results are generated, such as a high-risk early warning (High Risk): the deviation is greater than the preset allocation rate threshold and the load difference is greater than the load difference set value, indicating that a specific device is severely over-limited and the cluster allocation is extremely unbalanced, which requires immediate intervention; a medium-risk early warning (Medium Risk): the deviation is greater than the preset allocation rate threshold but the load difference does not exceed the load difference set value, indicating that a specific device is over-limited, but other devices still have surplus capacity, which can be considered for task transfer; a low-risk early warning / notice (Low Risk / Notice): the deviation is close to the preset allocation rate threshold or the load difference is large but does not exceed the load difference set value, prompting attention to potential risks or optimization opportunities for allocation; a balanced notice (Balanced Notice): the deviation is lower than the threshold and the load difference is lower than the load difference set value, indicating that the current allocation is good. Finally, the generated early warning results (including early warning levels, involved devices, deviations, load differences, and other key information) are sent in real time to a monitoring center large screen, a management personnel mobile terminal, or a field sound and light alarm device.

[0120] The present application can more comprehensively reflect the running state of hoisting equipment through comprehensive analysis of single-machine load allocation rate change gradient, load change gradient, and sliding window statistical characteristics, etc. multi-dimensional data, reducing the possible misjudgment of a single data dimension, and further improving the accuracy of early warning. The LSTM model with attention mechanism is used to process sequence data, improving the prediction accuracy. When it is predicted that a device may be overloaded (the prediction allocation rate exceeds the threshold), not only the over-limit degree (deviation) is calculated, but also the load balance of the entire cluster (load difference) is evaluated. Comprehensive analysis of the two generates early warning information of different levels, which can more comprehensively and intelligently perceive risks, and provides strong intelligent protection for the safe and efficient collaborative work of large hoisting equipment clusters.

[0121] In an embodiment, step S3 comprises:

[0122] According to the real-time pitch angle, a second collaborative work dynamic characteristic is determined, which is input into the hybrid anomaly detection model together with the operating condition data to obtain a first anomaly score corresponding to the isolation forest and a second anomaly score corresponding to the one-class support vector machine;

[0123] According to the work stage, a first weight and a second weight are determined in a preset weight library to weight and fuse the first abnormal score and the second abnormal score, to obtain a final abnormal score.

[0124] The final abnormal score is compared with a preset warning threshold, and a warning result is generated according to a comparison result to output.

[0125] Specifically, the torque data corresponding to the period of the real-time pitch angle is collected, so as to obtain a torque angle coordination coefficient by (torque A / torque B) / (pitch angle A / pitch angle B), which is used to measure the coordination consistency of the two devices in action, and the coefficient is taken as a second coordination work dynamic feature, and the feature and the operation condition data are input into a hybrid anomaly detection model for processing, in the hybrid anomaly detection model, an iForest model and an OC-SVM model are parallel and independent scoring; wherein the iForest model calculates an abnormal score Score_iForest according to the average path length of the data points in the random tree, that is, a first abnormal score; the OC-SVM calculates an abnormal score Score_OCSVM according to the distance of the data points to the decision boundary, that is, a second abnormal score.

[0126] According to the work stage, a first weight W_iForest and a second weight W_OCSVM are determined in a preset weight library to weight and fuse the first abnormal score and the second abnormal score, that is, Score_Final=W_iForest*Score_iForest+W_OCSVM*Score_OCSVM, to obtain a final abnormal score. Wherein, the weight library is configured as follows: in the flat hanging stage, W_iForest and W_OCSVM are 0.4 and 0.6 respectively; in the turning stage, W_iForest and W_OCSVM are 0.8 and 0.2 respectively; in the vertical stage, W_iForest and W_OCSVM are 0.3 and 0.7 respectively.

[0127] The final abnormal score is compared with a preset multi-level threshold (such as a prompt threshold, a warning threshold and an alarm threshold), and a warning of a corresponding level is triggered.

[0128] The application combines the dynamic features determined according to the real-time pitch angle and the operation condition data, uses the isolated forest to detect global outliers, uses the single-class support vector machine to be sensitive to local boundary anomalies, and the combination of the two can cover different types of anomalies and improve the detection robustness; the weights are adjusted according to the work stage, so that the evaluation result is more suitable for the actual working condition requirement; the final abnormal score obtained by the weighted fusion is compared with the threshold, so that the warning can be triggered in time to avoid the misjudgment or omission of a single model; the double-model score and the weight configuration are retained, which is convenient for subsequent analysis of the abnormal root cause, and also improves the warning accuracy.

[0129] In addition, it should be noted that the double-layer early warning model is trained separately, and the network parameters of the attention mechanism LSTM network are fixed when training the hybrid anomaly detection model, and the network parameters of the hybrid anomaly detection model are fixed when training the attention mechanism LSTM network. Other training methods can also be used. For the training of the hybrid anomaly detection model, a large amount of historical normal operation data (after the above feature engineering processing) is used to train the iForest model and the OC-SVM model, respectively. For the attention mechanism LSTM network, the historical multi-dimensional time series data (including all original and derived features) are constructed into fixed-length input sequences (such as using the past 60 seconds of data) and corresponding prediction targets for training, so that the model learns the time series relationship while learning how to allocate attention to different historical data points.

[0130] In another embodiment, the double-layer early warning model can also not perform early warning according to the business scenario:

[0131] The cooperative parameters output by the scenario simulation model under all business scenarios are combined with the operation condition data of each hoisting device to remove noise and outliers, and normalized to make different dimensional data comparable. After extracting the corresponding cooperative operation features such as load distribution rate change gradient, load change gradient, sliding window statistical features, and torque angle cooperative coefficient, they are input into the first layer of the hybrid anomaly detection model of integrated isolated forest and one-class support vector machine. This model detects outliers in the data by isolated forest (set an abnormal score threshold, and samples above the threshold are marked as suspicious samples), and uses one-class support vector machine to learn the distribution boundary of normal data (samples outside the boundary are determined as abnormal). It is determined whether the input data is abnormal. If both isolated forest and one-class SVM are marked as abnormal, the early warning is triggered directly. If only the single model alarms, it enters the second layer LSTM model for further verification. For data that needs to be detected again after the first layer, it is input into the LSTM network model with attention mechanism. This model can automatically learn the time series features and long-term dependencies in the data, and assign different weights to the data at different time steps through the attention mechanism to highlight important information. The model outputs the abnormal probability value of each hoisting device. If the output probability is greater than the preset probability threshold, the data is determined to be abnormal, and the difference between the output probability and the preset probability threshold is used to determine the early warning level.

[0132] The embodiment of the application designs a real-time early warning method for double-machine lifting based on how to realize real-time and accurate early warning of double-machine lifting. The method realizes real-time acquisition of lifting equipment operation condition data, dynamically updates a scene simulation model by combining digital twin technology, ensures that collaborative parameters are highly matched with actual working conditions, reduces collaborative failures caused by data lag, solves the problem of non-uniform data format by fusing heterogeneous data, improves model input quality, provides a reliable foundation for subsequent anomaly detection, quickly identifies anomalies by using a hybrid anomaly detection model (Isolation Forest + One-Class SVM), captures unknown anomalies with time sequence dependence by introducing an attention mechanism LSTM network, and realizes real-time and accurate early warning of double-machine lifting by digital twin technology and machine learning algorithms.

[0133] It should be noted that although each step in the above flowchart is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps has no strict order limitation, and these steps can be executed in other orders.

[0134] In another embodiment, as shown in FIG. 2, the second aspect of the application provides a real-time early warning system for double-machine lifting, comprising: Figure 2 A data processing module 10 is configured to acquire operation condition data of multiple lifting equipment in real time and perform standardized packaging to obtain heterogeneous processing data.

[0135] A model simulation module 20 is configured to construct a scene simulation model based on the heterogeneous processing data by digital twin technology to output collaborative parameters of each lifting equipment in multiple business scenarios.

[0136] A result output module 30 is configured to input each collaborative parameter and the operation condition data into a pre-constructed double-layer early warning model for processing to output an early warning result. The double-layer early warning model includes a hybrid anomaly detection model integrating Isolation Forest and One-Class Support Vector Machine, and an LSTM network model introducing an attention mechanism.

[0137] It should be noted that each module in the above real-time early warning system for double-machine lifting can be realized by software, hardware, and combinations thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to each module. For specific limitations of the real-time early warning system for double-machine lifting, refer to the limitations of the real-time early warning method for double-machine lifting in the above, which have the same functions and effects, and will not be described here.

[0138]

[0139] ​In conclusion, the present application relates to the field of information technology, and discloses a real-time early warning method and system for double-machine lifting, which comprises the following steps: acquiring running condition data of multiple lifting devices in real time and standardizing and packaging the data to obtain heterogeneous processing data; based on the heterogeneous processing data, constructing a scene simulation model through digital twinning technology to output coordination parameters of each lifting device in various business scenarios; inputting each coordination parameter and the running condition data into a pre-constructed double-layer early warning model for processing to output an early warning result; wherein the double-layer early warning model comprises a hybrid anomaly detection model integrating isolated forest and one-class support vector machine, and an LSTM network model introducing an attention mechanism; through digital twinning technology and machine learning algorithms, real-time and accurate early warning of double-machine lifting is realized.

[0140] Each embodiment in the specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment. It should be noted that, each technical feature of the above-mentioned embodiments can be combined arbitrarily, in order to make the description simple, not all possible combinations of each technical feature in the above-mentioned embodiments are described, however, as long as the combination of these technical features does not exist contradictory, it should be considered as the scope of the present application.

[0141] The above-mentioned embodiments only express several preferred embodiments of the present application, the description is more specific and detailed, but it should not be understood as the limitation of the patent scope of the present application. It should be pointed out that, for ordinary skilled in the art, without departing from the technical principles of the present application, several improvements and replacements can be made, and these improvements and replacements should be considered as the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the protection scope of the claims.

Claims

1. A real-time warning method for dual-machine lifting, characterized in that: include: Acquire the operating condition data of multiple lifting equipment in real time and perform standardized packaging to obtain heterogeneous processing data; Based on the heterogeneous processing data, a scenario simulation model is constructed through digital twin technology to output collaborative parameters of each of the lifting equipment in various business scenarios; The collaborative parameters and the operating condition data are input into a pre-built two-layer early warning model for processing, and an early warning result is output; wherein, the two-layer early warning model includes a hybrid anomaly detection model integrating isolation forest and single-class support vector machine and an LSTM network model introducing an attention mechanism.

2. A real-time warning method for dual-machine lifting according to claim 1, characterized in that: The real-time acquisition of the operating condition data of multiple hoisting equipment and the standardized packaging to obtain heterogeneous processing data include: Acquire operating condition data of multiple hoisting devices in real time, and determine the operating range of each hoisting device according to each operating condition data; Based on the operating condition data, the equipment performance library is called to match the rated lifting weight of each lifting equipment in the current posture in real time; The operating condition data, the operating ranges and the rated lifting weights are converted into a unified JSON format and timestamps are added to obtain heterogeneous processed data.

3. A real-time warning method for dual-machine lifting according to claim 1, characterized in that: The method of constructing a scene simulation model based on the heterogeneous processing data by using digital twin technology includes: Based on the heterogeneous processed data, each of the business scenarios is deconstructed to identify the physical entities of each business scenario, and the data sources of each physical entity are determined to construct a mapping table between the physical entities and the data sources; Performing three-dimensional geometric modeling on each of the hoisting devices according to the mapping table, and performing kinematic skeleton binding on the modeling results to generate a kinematic model of each of the hoisting devices; Introducing a virtual-real synchronization mechanism and a phase compensation coefficient to perform parameter mapping on each of the kinematic models and the heterogeneous processing data, obtaining a twin of each of the hoisting devices and performing multi-body collaborative aggregation to generate a twin aggregate model; simulating a test scenario using the twin polymer model to quantify an error between a simulation result and an actual result, and adjusting the phase compensation coefficient when the error does not meet a preset error; The steps of constructing the twin polymer model are iteratively executed according to the adjustment result until the obtained error meets the preset error, and the finally obtained twin polymer model is used as the scene simulation model.

4. A real-time warning method for dual-machine lifting according to claim 3, characterized in that: The three-dimensional geometric modeling of each hoisting device is performed according to the mapping table, and kinematic skeleton binding is performed on the modeling results to generate a kinematic model of each hoisting device, including: Constructing a three-dimensional geometric model of each of the hoisting devices through a three-dimensional modeling algorithm, and performing reproduction of key structures and environmental modeling to obtain a three-dimensional spatial structure model of each of the hoisting devices; Constructing an operational skeleton within each of the three-dimensional spatial structure models to define the connection relationship and operational degrees of freedom between the components of each of the hoisting devices, and generating an operational connection model for each of the hoisting devices; Based on the mapping table, the heterogeneous processed data is mapped to the skeletal parameters of each of the running connection models, and physical properties are assigned to obtain the kinematic model of each of the hoisting devices.

5. A real-time warning method for dual-machine lifting according to claim 3, characterized in that: The multi-body collaborative polymerization to generate a twin polymer model includes: Loading each of the twins into the same business scenario and performing absolute coordinate alignment on each of the twins; Establishing a multi-point constraint relationship between the hoisted object and each of the hoisting equipment according to the business logic of the business scenario; The collaborative behavior between the lifting devices under the multi-point constraint relationship is simulated through the physical engine and the script corresponding to the business logic to generate the twin aggregate model.

6. A real-time warning method for dual-machine lifting according to claim 5, characterized in that: The business scenario includes a joint lifting scenario, and the coordination parameters include a single-machine load distribution rate and a single-machine load; wherein, The output of the collaborative parameters of each hoisting equipment in a variety of business scenarios includes: The heterogeneous processing data is subjected to spatiotemporal synchronization calibration and then input into the scene simulation model for processing, so as to align the spatial coordinates of the hoisted object and each hoisting equipment in the common lifting scenario, and combine the physical engine to simulate the impact of single-machine action on the distribution rate, and output the single-machine load distribution rate and the single-machine load.

7. A real-time warning method for dual-machine lifting according to claim 5, characterized in that: The business scenario also includes a master-slave lifting scenario, and the collaborative parameters include the operation phase and the real-time pitch angle of the hoisted object; wherein, The output of the collaborative parameters of each hoisting equipment in a variety of business scenarios includes: The heterogeneous processing data is subjected to spatiotemporal synchronization calibration and then input into the scene simulation model for processing. Based on the business logic of the master-slave lifting scenario, each lifting device is associated with the inclination angle of the hoisted object through kinematic coupling, and the real-time pitch angle of the hoisted object is obtained and compared with the preset inclination angle threshold to determine the operation stage.

8. A real-time warning method for dual-machine lifting according to claim 6, characterized in that: The step of inputting the coordinated parameters and the operating condition data into a pre-built two-layer early warning model for processing and outputting an early warning result includes: determining a first collaborative operation dynamic feature based on the single machine load distribution rate and the single machine load, and constructing a sliding window statistical feature based on the operating condition data; Inputting the first collaborative operation dynamic features and the sliding window statistical features into the LSTM network model with the attention mechanism for processing, and outputting the single-machine load prediction distribution rate and the single-machine predicted load of each of the hoisting equipment; When the predicted distribution rate of the load of a single machine exceeds a preset distribution rate threshold, calculating the deviation between the predicted distribution rate of the load of a single machine and the preset distribution rate threshold, and quantifying the load difference between the predicted loads of each single machine; A warning result is generated and outputted based on the deviation amount and the load difference.

9. A real-time warning method for dual-machine lifting according to claim 7, characterized in that: The step of inputting the coordinated parameters and the operating condition data into a pre-built two-layer early warning model for processing and outputting an early warning result includes: Determining a second collaborative operation dynamic feature based on the real-time pitch angle, and inputting the feature and the operating condition data into the hybrid anomaly detection model for processing to obtain a first anomaly score corresponding to the isolation forest and a second anomaly score corresponding to the single-class support vector machine; Determining, according to the operation stage, a first weight and a second weight in a preset weight library, so as to perform weighted fusion on the first anomaly score and the second anomaly score to obtain a final anomaly score; The final abnormality score is compared with a preset warning threshold, and a warning result is generated and outputted based on the comparison result.

10. A real-time warning system for dual-machine lifting, characterized in that: include: The data processing module is used to obtain the operating condition data of multiple lifting equipment in real time and perform standardized packaging to obtain heterogeneous processed data; A model simulation module is used to construct a scenario simulation model based on the heterogeneous processing data through digital twin technology to output collaborative parameters of each of the lifting equipment in multiple business scenarios; The result output module is used to input each of the collaborative parameters and the operating condition data into a pre-built two-layer early warning model for processing and output the early warning result; wherein, the two-layer early warning model includes a hybrid anomaly detection model integrating isolation forest and single-class support vector machine and an LSTM network model introducing an attention mechanism.

Images