Grab bucket self-adaptive full-load control method and system based on no weight sensor

Abstract

The invention provides a grab bucket self-adaptive full-load control method and system based on no weight sensor, and relates to the technical field of crane control. According to the method, real-time data such as the voltage, the current, the rotating speed and the torque of a motor are collected, after filtering and normalization preprocessing, the data are input into a multi-layer feedforward neural network trained by a back propagation algorithm, and accurate grab bucket load identification under the sensorless condition is achieved. Meanwhile, a hybrid particle swarm optimization algorithm integrating gradient memory guidance, adaptive parameter adjustment and a stagnation particle restart mechanism dynamically adjusts the given torque of the motor and the lowering amount of the rope, and adaptive full load control in different goods operation scenes is ensured. Compared with a traditional control method, the method does not need an additional weight sensor, reduces the equipment cost and maintenance difficulty, and has a wide industrial application value. The technical problems that traditional port machine grab bucket control depends on artificial experience, an additional weight sensor is needed, and overload is easily caused by the change of the specific gravity of goods are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crane control, in particular to a grab bucket adaptive full load control method and system based on no weight sensor. BACKGROUND

[0002] In the process of port crane operation, the full load control precision of the grab bucket directly determines the operation efficiency and equipment safety. There are two core problems in the traditional crane grab bucket control scheme: first, the load detection relies on the weight sensor, and the sensor needs to be exposed to the harsh environment of port humidity, dust and strong vibration for a long time, which is easy to be disturbed and damaged, not only increases the equipment maintenance cost, but also causes control deviation due to inaccurate detection data; second, the control algorithm with fixed parameters is used, and the algorithm cannot adaptively adjust the control parameters, which is easy to cause the grab bucket overload or underload. Overload will cause the equipment structure stress to exceed the standard, which has safety hazards, and underload will reduce the operation efficiency, which is difficult to meet the modern port operation demand of high efficiency and safety.

[0003] Some existing improved schemes try to reduce the cost by simplifying the sensor structure, but they still rely on the sensor; another scheme optimizes the control algorithm, but it does not combine the load identification technology without sensor, and cannot fundamentally solve the defects of the traditional scheme.

[0004] Therefore, it is a key requirement in the current crane control field to develop a grab bucket full load control method without weight sensor and with adaptive parameter adjustment. SUMMARY

[0005] The purpose of the present application is to overcome the problems in the prior art, and to provide a grab bucket adaptive full load control method based on no weight sensor, and a system for implementing the method, which is suitable for the grab operation of different specific gravity and different types of cargo by the grab bucket of the crane in the port, wharf and other scenes, realizes accurate load identification and adaptive full load control without weight sensor, and improves the operation efficiency and safety performance.

[0006] In order to achieve the above purpose, the first aspect of the present application provides a grab bucket adaptive full load control method based on no weight sensor, which comprises the following steps: S1, collecting the motor operating parameters in real time through the multi-dimensional sensor group preset on the crane grab bucket driving motor; S2, constructing a load identification model based on deep feedforward neural network, inputting the standardized motor operating parameters in step S1 into the trained load identification model, and outputting the real-time load weight of the grab bucket; S3, a hybrid particle swarm optimization algorithm is constructed, the real-time load weight of the grab bucket output in step S2 is taken as an algorithm input, a minimum deviation of the grab bucket load from the target full load weight is taken as an optimization target, and optimal motor output torque and optimal wire rope length difference are solved; S4, the optimal motor output torque and the optimal wire rope length difference obtained in step S3 are converted into digital control signals recognizable by a PLC, the PLC drives the grab bucket driving motor to perform corresponding actions according to the digital control signals; in the process of operation of the grab bucket driving motor, motor operating parameters are collected in real time and fed back to the load identification model constructed in step S2, a new round of real-time load weight update is performed, and a closed loop feedback is formed; S5, the updated real-time load weight in step S4 is acquired, and a deviation from a preset target full load weight is calculated: If the real-time load weight is within a preset range of the target full load weight, it is determined that the grab bucket reaches a full load state, and a single grabbing operation is completed. If the real-time load weight is not within the preset range, return to step S3, take the updated real-time load weight as input again, restart the hybrid particle swarm optimization algorithm to perform parameter iterative adjustment until the full load determination condition is met.

[0007] As a preferred scheme of the first aspect, the motor operating parameters include: motor three-phase voltage, motor real-time speed, motor output load torque. The collected motor operating parameters are subjected to pre-processing operations including outlier rejection, data smoothing filtering and normalization processing, to obtain standardized motor operating parameters.

[0008] As a preferred scheme of the first aspect, the load identification model is constructed based on a deep feedforward neural network, and the number of input layer nodes of the load identification model is consistent with the dimension of the standardized motor operating parameters. The load identification model is trained by using an error back propagation algorithm, a training sample set includes matching data of standard operating parameters of the motor under different load weights and corresponding actual load weights, and the training is stopped when the prediction error of the load identification model is less than a predetermined proportion.

[0009] As a preferred scheme of the first aspect, the hybrid particle swarm optimization algorithm is constructed, specifically including: Record the historical iteration optimal gradient direction to make the particles converge quickly. Dynamically adjust the inertia weight and the learning factor according to the population diversity. Randomly reset the particles without improvement for continuous generations.

[0010] As a preferred scheme of the first aspect, the recording of the historical iteration optimal gradient direction to make the particles converge quickly specifically includes: Set population size Maximum number of iterations Location boundary , Velocity boundary Memory length Stagnation determination , ; Calculate the initial fitness of each particle. ;in Indicates the target full load weight; Indicates the initial position of the particle; This indicates the real-time load weight of the grab bucket corresponding to the initial position of the particle; Randomly initialize the position and velocity of each particle:

[0011]

[0012]

[0013]

[0014] set up ,set up , This represents the initial value of the particle's historical best fitness. Represents the initial value of the globally optimal fitness; gradient memoization queue Empty; stop counter for each particle ; For each particle Calculation via neural network And calculate fitness ;in This represents the real-time load weight of the grab bucket corresponding to the particle at the t-th iteration. This represents the fitness value of the particle at its current position in the t-th iteration (i.e., the deviation between the real-time load weight and the target full load weight at the t-th iteration). For each particle ,like ,but , Update the global optimum. ;in ; represents the globally optimal fitness value of a particle when it reaches the global optimum. Whenever a particle gains fitness, its position change is recorded in the gradient memory queue. ; If particles exist The position changes during the period and , add item to the gradient memory queue ; if , discard the oldest item; estimate the average direction with the gradient memory queue ; and ; calculate the gradient direction consistency: .

[0015] As a preferred solution of the first aspect, the dynamically adjusting the inertia weight and the learning factor according to the population diversity specifically comprises: inertia weight: ; wherein , ; respectively represent the maximum value of the inertia weight and the minimum value of the inertia weight diversity self-adaption: , ; wherein is the population mean, represents the distribution diversity index of the current position of the particle; learning factor: , ; wherein , respectively represent the individual learning factor and the social learning factor, , respectively represent the maximum value of the individual learning factor and the minimum value of the social learning factor, represents the iteration accuracy ratio, and the learning factor is linearly interpolated and adjusted according to the proportion of the current iteration number and the maximum iteration number; When is higher than a predetermined value, the gradient term will guide the velocity along the effective improvement direction and accelerate the local convergence; when is lower than a predetermined value, the gradient contribution is small, and the algorithm relies more on the global / individual term; for each particle, first calculate the standard PSO base velocity:

[0016] wherein represents the standard PSO base velocity of the particle at the t+1 iteration, represents the velocity of the particle at the t iteration; the gradient guided term is weighted and superimposed with the confidence: ; wherein is a gradient amplification coefficient, represents the gradient direction consistency; clip the velocity: , wherein denotes the maximum amplitude limit allowed for the velocity; The projected position is used for the next round of fitness calculation and gradient memory update, and the updated position is:

[0017] The position is projected: , ; wherein , respectively denote the constraint boundary after the particle update position.

[0018] As a preferred solution of the first aspect, the random resetting of the consecutive generations of particles without improvement specifically comprises: For each particle, the latest times of fitness change window is maintained, and the following is calculated: wherein denotes the fitness of the particle in the window, and if , the restart strategy is performed on the particle: perturbation restart: ; wherein denotes the perturbation coefficient (i.e. the random perturbation range near the global optimal position); hybrid restart: wherein , are the positions of two excellent individuals in the current population; velocity resetting: , and the stagnation counter and gradient memory are updated.

[0019] As a preferred solution of the first aspect, the minimum deviation of the grab load from the target full load weight is taken as the optimization target, and the optimal motor output torque and optimal wire rope length difference are solved and output, specifically comprising: if , is an error threshold, then is output, wherein , are the optimal motor output torque and the optimal wire rope length difference, respectively.

[0020] The second aspect of the application provides a grab self-adaptive full load control system based on a weight sensor, which is used to automatically execute the grab self-adaptive full load control method based on a weight sensor of the first aspect. The system comprises: A multi-dimensional sensor group is arranged on the grab driving motor of the crane; the multi-dimensional sensor group is used to collect motor operating parameters and transmit them to the edge controller at a predetermined sampling frequency; The edge controller, as a data interaction core node, realizes multi-module data transfer and preprocessing; the edge controller is used for executing preprocessing on motor operation parameters and realizing bidirectional communication: on one hand, receiving a target full load weight setting instruction sent by a client and forwarding the target full load weight setting instruction to a server; on the other hand, forwarding optimal motor torque and optimal rope difference parameters output by the server to a PLC motion control module; The client, as a man-machine interaction interface, can set parameters and monitor states, and displays a current load weight of a grab bucket, motor operation parameters, a PLC control state and an error range in real time; when a load weight deviation exceeds a preset safety threshold or equipment is abnormal, sound and light alarms are triggered and are synchronized to the server; The server, as an algorithm execution core, realizes load identification and parameter optimization; The PLC motion control module, as an execution terminal, reads optimal motor torque and optimal rope difference parameters from the edge controller, controls a motor driver to adjust torque of the motor and adjust a rope difference, and realizes accurate control of opening and closing degrees and grabbing force of the grab bucket.

[0021] As a preferred scheme of the second aspect, the server comprises a neural network load identification unit and a full load control algorithm operation unit; The neural network load identification unit is used for loading a pre-trained deep feedforward neural network, receiving normalized real-time parameters transmitted by the edge controller, and outputting real-time load weight of the grab bucket; The full load control algorithm operation unit is used for inputting the real-time load weight into a hybrid particle swarm optimization algorithm, and calculating optimal motor torque and optimal rope difference.

[0022] Beneficial effects: The application solves the technical problems of traditional port machinery grab bucket control, such as dependence on manual experience, need for additional weight sensors, and easy overload caused by changes in specific gravity of goods. The method collects real-time data such as voltage, current, speed and torque of the motor, pre-processes the data through filtering and normalization, inputs the data into a multi-layer feedforward neural network trained by a back propagation algorithm, realizes accurate identification of the load of the grab bucket under the condition of no sensor, and dynamically adjusts the given torque of the motor and the amount of rope drop through a hybrid particle swarm optimization algorithm combining gradient memory guidance, adaptive parameter adjustment and stagnation particle restart mechanism, to ensure adaptive full load control in different goods operation scenarios. Compared with the traditional control method, the application does not need additional weight sensors, reduces equipment cost and maintenance difficulty, and has wide industrial application value. The application has more accurate efficiency and more extensive flexibility. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the grab bucket adaptive full load control method without weight sensors.

[0024] Figure 2A schematic diagram of the neural network model for the experiment of the present application.

[0025] Figure 3 A flow chart of the load identification of the neural network for the experiment of the present application.

[0026] Figure 4 A flow chart of the hybrid particle swarm optimization algorithm for the experiment of the present application.

[0027] Figure 5 A comparison chart of the adaptive full-load control effect of the present application. DETAILED DESCRIPTION

[0028] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without one or more of these specific details. In other instances, well-known features have not been described in detail in order to avoid unnecessarily complicating the present application.

[0029] The embodiment discloses a grab bucket adaptive full-load control method based on a weight sensor, as shown in the figure, the steps are: Figure 1 S1: Real-time collection of motor operating parameters through a multi-dimensional sensor group preset on the crane grab bucket drive motor; the operating parameters at least include motor three-phase voltage, motor real-time speed, motor output load torque; pre-processing of the collected operating parameters, the pre-processing includes outlier rejection, data smoothing filtering, normalization processing, to obtain standardized real-time operating parameters.

[0030] S2: Construction of a load identification model based on a deep feedforward neural network (BP neural network), the number of input layer nodes of the model is consistent with the dimension of the pre-processed operating parameters. The model is trained using the error back propagation algorithm, the training sample set includes matching data of the standard operating parameters of the motor under different load weights and the corresponding actual load weights, and the training is stopped when the model prediction error is less than 5%; input the standardized real-time operating parameters in S1 into the trained neural network load identification model, and output the real-time load weight of the grab bucket.

[0031] S3: Construction of a hybrid particle swarm optimization algorithm, the algorithm integrates three core mechanisms: ① gradient memory guidance mechanism (records the historical optimal gradient direction of iteration, so that the particles converge quickly); ② adaptive parameter adjustment mechanism (dynamically adjusts the inertia weight and learning factor according to the population diversity); ③ stagnation particle restart mechanism (randomly resets the particles that have no improvement for continuous generations to avoid falling into local optimum); input the real-time load weight output by S2 into the algorithm, and take "the minimum deviation of the grab bucket load and the target full-load weight" as the optimization goal, to solve the optimal motor output torque and the optimal wire rope length difference (rope difference).​

[0032] S4: Convert the optimal motor torque and optimal rope difference obtained in S3 into a digital control signal recognizable by PLC and input into the control system; PLC drives the grab bucket driving motor to perform corresponding actions (adjusts the motor output torque and controls the steel wire rope length) according to the control signal, so as to realize precise control of the grab bucket load; at the same time, in the process of motor operation, the corresponding motor operating parameters at this time (the same as the parameter types in S1) are collected in real time and transmitted to the neural network load identification model of S2 for a new round of real-time load weight update, forming a closed loop feedback.

[0033] S5: Calculate the deviation of the real-time load weight updated by the neural network model in S4 from the preset target full load weight; if the real-time load weight is within ±2.4% of the target full load weight, it is determined that the grab bucket reaches the full load state and completes a single grabbing operation; if the real-time load weight is not within the above range, return to S3, input the updated real-time load weight, restart the hybrid particle swarm optimization algorithm for parameter iteration adjustment until the full load determination condition is met.

[0034] In S1 above, the collected motor parameters are normalized to eliminate the influence of dimensional differences on the neural network model. The normalization formula is as follows: ; wherein, is the original parameter ( / / ), , are the historical minimum and maximum values of the parameter, is the normalized parameter.

[0035] In S2 above, as shown in Figure 2 Figure 3 , a neural network model is constructed to identify the grab bucket load weight (output) based on the normalized motor parameters (input).

[0036] In S3 above, as shown in Figure 4 , the real-time load weight is input into the hybrid particle swarm optimization algorithm, which combines gradient memory guidance, adaptive parameter adjustment, and stagnation particle restart mechanism to output the optimal support rope torque and optimal rope difference. The specific steps are as follows: S3-1: initialization stage, allocate data structure for gradient memory and stagnation counter for subsequent mechanisms. Set the population size ; the maximum number of iterations ; the position boundary , ; the speed boundary Memory length Stagnation determination , ; Calculate the initial fitness of each particle Randomly initialize the position and velocity of each particle: , , , ;set up ,set up Gradient memory queue Empty; stop counter for each particle

[0037] S3-2: Gradient memory requires history Corresponding The gradient is estimated using values; adaptive parameters can be calculated based on population diversity (see below), and stagnation determination uses fitness time series. For each particle... Calculation via neural network (Should (Predicted weight at each moment), and calculate fitness. .

[0038] S3-3: When a particle becomes the new global optimum, the gradient memory queue should record the particle's position change and fitness improvement to provide samples for subsequent gradient estimation. For each particle, if (its historical individual optimal fitness), then , Update the global optimum. .

[0039] S3-4: This step is the core of gradient memory guidance: when A high gradient indicates consistent historical improvement direction and reliable gradient terms; a low gradient suppresses its influence and avoids noise-induced errors. Whenever a particle experiences a fitness improvement (especially a global or individual optimal update), its position change is recorded in the gradient memory queue. .

[0040] If particles exist The position changes during the period and Then add item arrive ;like Then discard the oldest item.

[0041] use Estimating the mean direction (approximate gradient direction) using mid-data:

[0042] And calculate the gradient direction consistency (confidence):

[0043] S3-5: Adaptive parameter adjustment takes effect: early Large, Large (encourage individual exploration); later Small, Large (encourage group convergence); if group diversity Small (group has focused), can be temporarily increased Or trigger partial restart to prevent premature convergence.

[0044] Inertia weight:

[0045] Diversity adaptation: ,

[0046] Learning factor: ,

[0047] S3-6: When High (historical direction consistency), gradient term will guide velocity in the direction of effective improvement, accelerating local convergence; when Low, gradient contribution is small, algorithm relies more on global / individual terms, maintaining exploratory nature. For each particle, first calculate the standard PSO base velocity:

[0048] Add the gradient guidance term with confidence weighting: , where is the gradient amplification coefficient.

[0049] Clip the velocity (prevent step size from being too large):

[0050] S3-7: Projected position is used for next round of fitness calculation and gradient memory update Update position:

[0051] Project the position (ensure physical constraints): ,

[0052] S3-8: Stagnation restart directly solves local optimal problem: when a particle has long-term improvement, restart injects a new direction or "grafts" it to the combination of excellent individuals, improving group diversity and global exploration ability; combined with adaptive parameters: if the overall group diversity is small and frequent stagnation, the strategy can temporarily increase Or increase the restart probability.

[0053] Maintain the latest times of fitness change window for each particle, calculate: , if (particle stagnation), then perform a restart strategy on the particle: perturbation restart:

[0054] hybrid restart: where , are the positions of the two best individuals in the current population.

[0055] velocity reset: , update the stagnation counter and gradient memory S3-9: Check termination condition at each generation: if (error threshold), then output where , are the optimal motor output torque and optimal wire length difference (wire difference), respectively.

[0056] Next, taking the port crane grab operation as the experimental scene, the driving motor is a three-phase asynchronous motor. Through the edge controller, the data is interacted in real time, and the collected parameters include motor voltage, speed, and torque. Through the standard weighing sensor (accuracy 0.1%), the load weight is calibrated, and a training data set containing 5000 groups of samples (covering 500-4500kg) is constructed. In this way, the effectiveness and universal adaptability of the method proposed in this embodiment are verified.

[0057] As shown in Figure 5 , after starting the crane control system, loading the trained neural network model and hybrid PSO algorithm parameters, data collection is performed. For a target weight of 4200kg, the system outputs the optimal torque of 35.4N·m and the optimal wire difference of 161.7cm after 4 iterations. The full load error is within 2.4%, meeting the requirements, stopping adjustment, and completing a single operation. The entire operation process has no manual intervention, the accuracy and efficiency meet the design goals, and there is no overload or underload in the continuous operation test.

[0058] In another embodiment, the present application discloses a grab bucket adaptive full load control system based on no weight sensor. The control system includes a data acquisition module, an edge controller, a client, a server and a PLC motion control module, each module forms a closed loop control through data interaction, and the specific cooperative modules are as follows: (1) Data acquisition module: used for collecting real-time running parameters of the crane grab driving motor, the acquisition module transmits the original parameters to the edge controller at a sampling frequency of 100-200Hz.

[0059] (2) Edge controller: as the core node of data interaction, it realizes multi-module data transfer and preprocessing. It performs normalization processing on the original parameters transmitted by the acquisition module, and eliminates abnormal values (such as parameters exceeding the rated range of the device, signal mutation values); and realizes bidirectional communication, on the one hand, it receives the target full load weight setting instruction sent by the client (such as the user inputting the rated full load weight of the grab bucket through the client), and forwards it to the server; on the other hand, it forwards the optimal motor torque and optimal rope difference parameters output by the server to the PLC motion control module.

[0060] (3) Client: as a human-computer interaction interface, it can set parameters. It supports user input of target full load weight and job type. It can also monitor the state, and display the current load weight of the grab bucket (server identification result), motor operating parameters, PLC control state and error range in real time. When the load weight deviation exceeds the preset safety threshold or the device is abnormal, sound and light alarms are triggered and synchronized to the server.

[0061] (4) Server: as the core of algorithm execution, it realizes load identification and parameter optimization, including the following specific functions: Neural network load identification: load a pre-trained multi-layer feedforward neural network model, receive real-time parameters transmitted by the edge controller after normalization, and output the real-time load weight of the grab bucket; Full load control algorithm operation: input the real-time load weight into the hybrid particle swarm optimization algorithm (combining gradient memory guidance, adaptive parameter adjustment, and stagnation particle restart mechanism), calculate the optimal motor torque and optimal rope difference; (5) PLC motion control module: as the execution terminal, it realizes closed-loop control of the grab bucket action, including the following specific functions: Parameter receiving and execution: read the optimal motor torque and optimal rope difference parameters from the edge controller, control the motor driver to adjust the torque of the motor and adjust the rope difference, and realize accurate control of the opening and closing degree and grabbing force of the grab bucket.

[0062] In another embodiment, the present application provides a computer readable storage medium storing a computer program, which causes a computer to execute the grab bucket adaptive full load control method based on no weight sensor of the foregoing embodiment.

[0063] In another embodiment, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the grab bucket adaptive full load control method based on no weight sensor of the foregoing embodiment when executing the computer program.

[0064] In embodiments disclosed herein, a computer storage medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. Computer storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media can include, but are not limited to, a one or more lines of electrical connections, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0065] As mentioned above, the application has been described and expressed with reference to particular preferred embodiments, but it is not to be construed as being in any way limited thereto or as being limited by the principles and features as set forth in the claims. Various modifications in form and details can be made without departing from the spirit and scope of the application as defined in the claims.

Claims

1. A grab bucket adaptive full-load control method based on a weightless sensor, characterized in that, Includes the following steps: S1. Real-time acquisition of motor operating parameters through a multi-dimensional sensor group pre-installed on the crane grab bucket drive motor; S2. Construct a load identification model based on a deep feedforward neural network. Input the standardized motor operating parameters from step S1 into the trained load identification model and output the real-time load weight of the grab bucket. S3. Construct a hybrid particle swarm optimization algorithm. Take the real-time load weight of the grab bucket output in step S2 as the algorithm input. Take the minimum deviation between the grab bucket load and the target full load weight as the optimization objective, and solve for the optimal motor output torque and the optimal wire rope length difference. S4. The optimal motor output torque and the optimal wire rope length difference obtained in step S3 are converted into digital control signals that can be recognized by the PLC. The PLC drives the grab bucket drive motor to perform corresponding actions according to the digital control signals. During the operation of the grab bucket drive motor, the motor operation parameters are collected in real time and fed back to the load identification model constructed in step S2 to perform a new round of real-time load weight update, forming a closed-loop feedback. S5. Obtain the updated real-time load weight from step S4 and calculate the deviation from the preset target full load weight: If the real-time load weight is within the preset range of the target full load weight, it is determined that the grab has reached the full load state and the single grab operation is completed. If the real-time load weight is not within the preset range, return to step S3, use the updated real-time load weight as input again, and restart the hybrid particle swarm optimization algorithm to iteratively adjust the parameters until the full load determination condition is met.

2. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 1, characterized in that, The motor operating parameters include: motor three-phase voltage, motor real-time speed, and motor output load torque; The collected motor operating parameters are preprocessed, including outlier removal, data smoothing and filtering, and normalization, to obtain standardized motor operating parameters.

3. The adaptive full-load control method for a grab bucket based on a weightless sensor according to claim 1, characterized in that, The load identification model is constructed based on a deep feedforward neural network, and the number of nodes in the input layer of the load identification model is consistent with the dimension of the standardized motor operating parameters. The load identification model is trained using the backpropagation algorithm. The training sample set includes the standard operating parameters of the motor under different load weights and the matching data of the corresponding actual load weights. Training stops when the prediction error of the load identification model is less than a predetermined proportion.

4. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 1, characterized in that, The construction of the hybrid particle swarm optimization algorithm specifically includes: Record the optimal gradient direction in the historical iterations to enable particles to converge quickly; The inertia weight and learning factor are dynamically adjusted based on population diversity. Randomly reset particles that have not been improved for multiple consecutive generations.

5. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 4, characterized in that, The recording of the optimal gradient direction in historical iterations, enabling particles to converge quickly, specifically includes: Set population size Maximum number of iterations Location boundary , Velocity boundary Memory length Stagnation determination , ; Calculate the initial fitness of each particle. ;in Indicates the target full load weight Indicates the initial position of the particle; This indicates the real-time load weight of the grab bucket corresponding to the initial position of the particle; Randomly initialize the position and velocity of each particle: set up ,set up , This represents the initial value of the particle's historical best fitness. Represents the initial value of the globally optimal fitness; gradient memoization queue Empty; stop counter for each particle ; For each particle Calculation via neural network And calculate fitness ;in This represents the real-time load weight of the grab bucket corresponding to the particle at the t-th iteration. This represents the fitness value of the particle at its current position in the t-th iteration, which is the deviation between the real-time load weight and the target full load weight at the t-th iteration. For each particle ,like ,but , Update the global optimum. ;in This represents the globally optimal fitness value of a particle when it is in the global optimum. Whenever a particle gains fitness, its position change is recorded in the gradient memory queue. ; If particles exist The position changes during the period and Then add item to gradient memory queue ;like Then discard the oldest item; Gradient memory queue Mean direction of data estimation: ; Consistency in gradient direction calculation: .

6. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 4, characterized in that, The dynamic adjustment of inertia weights and learning factors based on population diversity specifically includes: Inertia weight: ;in , These represent the maximum and minimum values ​​of the inertial weight, respectively. Diversity Adaptation: , ;in The group mean An index representing the diversity of particle distribution at its current location; Learning factors: , ;in , These represent individual learning factors and social learning factors, respectively. , These represent the maximum value of the individual learning factor and the minimum value of the social learning factor, respectively. This represents the iteration accuracy ratio, which is used to linearly interpolate and adjust the learning factor based on the ratio of the current iteration number to the maximum iteration number. when Above the predetermined value, the gradient term will guide the velocity along the effective improvement direction, accelerating local convergence; when Below the predetermined value, the gradient contribution is small, and the algorithm relies more on global / individual terms; for each particle, the standard PSO basis velocity is calculated first: in This represents the standard PSO basis velocity of the particle in the (t+1)th iteration. This represents the velocity of the particle in the t-th iteration; The gradient-guided terms are weighted and summed using confidence levels: ;in This is the gradient amplification factor. This indicates consistency in gradient direction; Trim the speed: ,in Indicates the maximum permissible speed limit; The projected position is used for the next round of fitness calculation and gradient memory update. Update position: Projection of position: 、 ； in , These represent the constraint boundaries after the particle's position is updated.

7. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 4, characterized in that, The random reset of particles that have not improved for multiple consecutive generations specifically includes: Maintain the latest for each particle Calculate the fitness change window: ,in This indicates the fitness of the particles within the window. Then, a restart strategy is executed on the particles: Disturbance restart: ;in Indicates the disturbance coefficient; Hybridization Reboot: ,in , This refers to the positions of two outstanding individuals within the current group; Speed ​​Reset: Update the stagnation counter and gradient memory.

8. The grab bucket adaptive full-load control method based on a weightless sensor according to claim 7, characterized in that, With the goal of minimizing the deviation between the grab bucket load and the target full-load weight, the optimal motor output torque and the optimal wire rope length difference are determined, specifically including: like , If the error threshold is used, then the output will be... ,in , These represent the difference between the optimal motor output torque and the optimal wire rope length, respectively.

9. A grab bucket adaptive full-load control system based on a weightless sensor, used to automatically execute the grab bucket adaptive full-load control method based on a weightless sensor as described in any one of claims 1 to 8, characterized in that, include: A multi-dimensional sensor array is deployed on the crane grab bucket drive motor; The multi-dimensional sensor group is used to collect motor operating parameters and transmit them to the edge controller at a predetermined sampling frequency; The edge controller, as the core node for data interaction, realizes data transfer and preprocessing between multiple modules. The edge controller is used to perform preprocessing on motor operating parameters and realize bidirectional communication: on the one hand, it receives the target full load weight setting command sent by the client and forwards it to the server; on the other hand, it forwards the optimal motor torque and optimal rope difference parameters output by the server to the PLC motion control module. The client, serving as a human-machine interface, allows for parameter setting and status monitoring. It displays the current load weight of the grab bucket, motor operating parameters, PLC control status, and error range in real time. When the load weight deviation exceeds the preset safety threshold or the equipment malfunctions, it triggers an audible and visual alarm and synchronizes it to the server. The server-side, as the core of algorithm execution, implements load identification and parameter optimization; The PLC motion control module, as the execution terminal, reads the optimal motor torque and optimal rope difference parameters from the edge controller, controls the motor driver to adjust the motor torque and rope difference, and achieves precise control of the grab opening and closing degree and grabbing force.

10. A grab bucket adaptive full-load control system based on a weightless sensor according to claim 9, characterized in that, The server includes a neural network load identification unit and a full load control algorithm calculation unit; The neural network load identification unit is used to load the pre-trained deep feedforward neural network, receive the normalized real-time parameters transmitted by the edge controller, and output the real-time load weight of the grab bucket. The full-load control algorithm calculation unit is used to input the real-time load weight into the hybrid particle swarm optimization algorithm to calculate the optimal motor torque and the optimal rope difference.

Images