Electric hoist torque allocation method and system for coping with load mutation working conditions

Abstract

The present application belongs to the technical field of intelligent control of hoisting equipment, and particularly relates to a method and system for adjusting the torque of an electric hoist in response to sudden changes in load conditions. The method comprises: generating a comprehensive confidence level for sudden changes in load conditions through multi-source signal fusion and confirming the sudden change conditions; outputting a future torque with a probability weight and a derivative quantity prediction sequence based on an adaptive scene tree; generating a change rate constraint mask, a vibration constraint mask, and a thermal safety mask based on the temperature, speed, and upper limit characteristics of multiple execution units, and fusing them into a comprehensive constraint mask; then obtaining the torque working domain of each execution unit within the predicted time domain; using rule heuristics and evolutionary optimization in parallel to solve and select the feedforward torque command after arbitration; and using the actual torque deviation to correct the feedforward gain coefficient in real time, thereby achieving high-robustness sudden change recognition, uncertain torque multi-path prediction, dynamic working domain-based collaborative adjustment, and online gain calibration optimization.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for lifting equipment, specifically relating to a method and system for adjusting the torque of an electric hoist to cope with sudden load changes. Background Technology

[0002] As a typical material handling equipment, electric hoists face increasingly complex load conditions in construction, manufacturing, and logistics scenarios. During actual lifting and translation, events such as sudden slippage of the hoisted object, instantaneous tensioning after slack in the slings, and the traveling mechanism passing through track joints can all trigger sudden load changes on the order of milliseconds to seconds. The impact torque generated by these sudden conditions not only worsens the vibration amplitude at the lifting end, reducing positioning accuracy and safety, but also creates short-term overload hotspots within the transmission chain and actuators, threatening the fatigue life of the motor insulation and reduction mechanism. Traditional electric hoists with power frequency drive or simple variable frequency speed regulation often use fixed-proportion or steady-state model-based PID control for torque distribution. These methods suffer from sluggish response and limited compensation during sudden changes, making it difficult to achieve coordinated output among multiple actuators while ensuring system stability.

[0003] To address these issues, the industry has introduced control architectures based on load observers or disturbance feedforward. Some solutions estimate load torque using Kalman filters or sliding mode observers, and compensate using single-step or torque motors, but these can only cover disturbances in specific frequency bands. Other solutions establish coarse operating condition identification models, triggering preset allocation strategies based on current or torque thresholds, but these thresholds are fixed and cannot adapt to changing working environments and equipment states. Furthermore, existing technologies generally lack parallel constraints on torque change rate, vibration intensity, and motor thermal state. Especially when multiple execution units coexist, they cannot provide a torque operating domain that balances dynamic constraints and efficiency optimization within the prediction time domain, leading to frequent over-limit alarms under abrupt operating conditions, premature protective derating intervention, and even misjudged shutdowns, severely impacting production efficiency and equipment availability. Summary of the Invention

[0004] To overcome the problems of insufficient accuracy in load mutation identification, high uncertainty in torque prediction, lack of parallel fusion of multiple constraints, and inability to balance tracking error and vibration suppression in torque allocation in existing technologies, this invention provides a torque allocation method and system for electric hoists to cope with load mutation conditions. It achieves highly robust mutation identification, uncertainty-adaptive torque multi-path prediction, collaborative allocation based on dynamic working domain, and closed-loop optimization with online gain calibration. This significantly reduces torque tracking error and hoist end vibration under sudden shocks, while preventing thermal overload of the actuator and excessive torque change rate.

[0005] The technical solution of this application specifically includes: According to one aspect of this application, a method for adjusting the torque of an electric hoist to cope with sudden load changes is provided, comprising: During the operation of the electric hoist, the stator current, rotor speed, drum end torque and lifting end acceleration are collected. The transient power fluctuation component is extracted by the stator current and rotor speed and the anomaly degree is calculated. The real-time torque change rate and real-time vibration intensity are calculated by the drum end torque and lifting end acceleration, respectively. The load mutation comprehensive confidence degree is generated by weighted fusion. When the load mutation comprehensive confidence degree exceeds the threshold and the speed drop exceeds the tolerance, it is determined that the load mutation condition has been entered. For load change conditions, a multi-branch probability scenario tree is constructed and the overall prediction uncertainty is calculated. When the time limit is exceeded, the bandpass filter parameters are adjusted and corrected again, and the future torque prediction sequence, the future torque change rate prediction sequence, and the future vibration intensity prediction sequence are output. The upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity are obtained for the high-inertia main drive motor and the auxiliary torque compensation device. The future torque change rate prediction sequence and the future vibration intensity prediction sequence are compared with the corresponding upper limit values ​​to generate the change rate constraint mask and the vibration constraint mask. The thermal safety mask calibrated by the upper limit of the basic torque is fused into a comprehensive constraint mask to determine the torque working domain. The future torque prediction sequence with the highest cumulative probability is separated into low-frequency base load components and high-frequency residual components. The instruction sequence is generated in parallel by a rule-based heuristic solver and an evolutionary optimization solver. The better one is selected as the feedforward torque instruction by online arbitration. During the execution of the feedforward instruction, the actual output torque is collected and compared with the feedforward torque instruction to obtain the torque deviation sequence. The feedforward gain coefficient shared by the rule-based heuristic solver and the evolutionary optimization solver is corrected in real time based on the torque deviation sequence, so that the instruction is more in line with the actual dynamic characteristics of the execution unit.

[0006] Preferably, the extraction of transient power fluctuation components and the calculation of anomaly include: Instantaneous active power is calculated using stator current and reconstructed stator voltage. Steady-state power components are eliminated by a high-pass filter to obtain the transient power signal. The transient power signal is fed into a bandpass filter with adjustable center frequency and bandwidth to extract the power fluctuation component sequence. ; The effective value of the power fluctuation component is calculated using a sliding window and compared with the reference steady-state effective value. Compare and calculate the abnormality of load mutation according to the following formula. : In the formula, The length of the sliding window. These are the effective values ​​of the fluctuation components under the same window length during the no-load steady-state operation phase. This is the sequence number of the current control cycle.

[0007] Preferably, the calculation of the real-time torque change rate and real-time vibration intensity includes: The first derivative of the torque is calculated by numerical differentiation of the discrete torque sequence at the drum end, and the differentiation result is low-pass filtered to obtain the real-time torque change rate. ; The acceleration signal at the end of the spreader, after removing the gravity bias, is calculated using the following formula. Real-time vibration intensity per control cycle : In the formula, This represents the number of sampling points corresponding to the short-time energy integral window width. This is the vertical acceleration after removing the gravitational bias.

[0008] Preferably, the method for determining the load change condition includes: abnormality The real-time torque change rate amplitude and real-time vibration intensity are normalized and mapped to the [0,1] interval to obtain the normalized anomaly degree. Normalized real-time torque change rate and normalized real-time vibration intensity ; According to preset weighting coefficients Perform weighted fusion to generate a comprehensive confidence score for load mutation. : ; Calculate the drop in current rotor speed relative to the short-term historical average. When the drop is continuous The control cycle exceeds the speed drop threshold, and If the preset confidence threshold is exceeded, the electric hoist is determined to have entered a sudden load change condition.

[0009] Preferably, the multi-branch probability scene tree construction step includes: Historical torque trajectories were statistically analyzed for each of the lifting, translation, and descent phases. The torque value range was discretized into multiple state intervals. The state transition count matrix was statistically analyzed and normalized to obtain the basic state transition probability matrix. ; Using the normalized real-time torque change rate and normalized real-time vibration intensity The base transition probability is corrected to obtain the corrected transition probability. : In the formula, This represents the current torque state. This represents the possible torque state at the next moment. and As a preset impact factor, The indicator function represents the possible torque state at the next moment. The value is 1 if the direction of the corresponding torque change is consistent with the direction of the real-time torque change rate; otherwise, it is 0. Using the current torque state as the root node, the tree expands outward layer by layer according to the corrected transition probability to the preset prediction steps, thus constructing a multi-branch probability scenario tree.

[0010] Preferably, the multi-branch probability scene tree construction step further includes: Reconstruct the future torque prediction sequence based on the center value of the torque state interval at each step on each path in the scene tree; By performing forward time difference on the future torque prediction sequence, the future torque change rate prediction sequence is obtained; Based on the statistical mapping relationship between the rate of change of torque and vibration intensity, the future vibration intensity prediction sequence is calculated. The probability of each path is output as the probability weight of the corresponding prediction sequence.

[0011] Preferably, the calculation of the overall prediction uncertainty and the readjustment of the bandpass filter parameters when exceeding the limit include: Collect the path probabilities of all complete paths reaching the predicted depth in the scene tree. Calculate the overall prediction uncertainty using the following formula. : ; when Exceeding the allowed limit At that time, along with The center frequency and bandwidth of the bandpass filter are iteratively adjusted in the descent gradient direction. The transient power fluctuation components, anomalies, real-time torque change rate and real-time vibration intensity are re-extracted, and the scenario tree branch probability is corrected until the overall prediction uncertainty converges to below the allowable upper limit or reaches the maximum number of iterations.

[0012] Preferably, the steps for generating the rate of change constraint mask, vibration constraint mask, and thermal safety mask include: The absolute value of the predicted future torque change rate sequence is taken by weighting the path probability and comparing it with the upper limit of the torque change rate of each execution unit to generate a change rate constraint mask. The predicted sequence of future vibration intensity is weighted by path probability, and a preset vibration sensitivity amplification factor is applied to the auxiliary torque compensation device. The result is then compared with the upper limit value of vibration intensity for each execution unit to generate a vibration constraint mask. Based on the preset temperature-torque upper limit derating relationship, the allowable torque upper limit corresponding to the currently collected execution unit temperature is compared with the basic torque upper limit to calibrate the thermal safety mask.

[0013] Preferably, the step of determining the torque operating domain using the comprehensive constraint mask includes: Rate of change constraint mask Vibration constraint mask thermal security mask Perform a logical OR operation element-wise on the prediction step and each cell to obtain the synthesized constraint mask. : In the formula Indicates the execution unit identifier. Indicates the prediction step. Indicates the logical "OR"; when When the indication is limited, the corresponding unit in the prediction step is determined according to the dominant constraint type. The upper limit of the torque is reduced to a value that satisfies the corresponding constraints, forming the torque working domain. .

[0014] Preferably, the instruction sequence generation step includes: The future torque prediction sequence with the highest cumulative probability is smoothed by using a moving average filter to separate the low-frequency base charge component and the high-frequency residual component. The operating rule-based heuristic solver prioritizes the allocation of low-frequency base load components to the high-inertia main drive motor within the torque operating domain, allocates high-frequency residual components to the auxiliary torque compensation device, and performs proportional reduction on any out-of-limit parts to generate a main control command sequence. Simultaneously, an evolutionary optimization solver is run to search for Pareto front solutions within the same torque operating domain, aiming to minimize the rate of change of instructions, maximize system efficiency, and minimize over-limit penalties, thereby generating alternative instruction sequences.

[0015] Preferably, the feedforward torque command selection step includes: The system calls upon the real-time collected torque signals at the drum end and acceleration signals at the spreader end to perform transient response prediction on the main control command sequence and alternative command sequences, and calculates the weighted arbitration index of torque following error and vibration intensity response. The set of instructions that minimizes the weighted arbitration index is selected as the final feedforward torque command and sent to the high-inertia main drive motor and auxiliary torque compensation device.

[0016] Preferably, the method for obtaining the torque deviation sequence includes: The actual electromagnetic torque of the high-inertia main drive motor is calculated by the torque current and torque constant output by the driver, and the actual output torque of the auxiliary torque compensation device is estimated by the torque sensor or servo motor current. In each control cycle, the difference between the feedforward torque command and the actual output torque is calculated point by point to obtain the torque deviation sequence and store it in the sliding window.

[0017] Preferably, the feedforward gain coefficient shared by the real-time correction rule heuristic solver and the evolutionary optimization solver based on the torque deviation sequence includes: A feedforward channel gain mismatch model is established, and the actual output torque is expressed as the product of the feedforward gain coefficient and the feedforward torque command. The recursive least squares method with a forgetting factor is used to estimate the feedforward gain coefficient online using the torque deviation sequence; After limiting the estimated value to a preset range, it is synchronously updated to the instruction generation model of the rule-based heuristic solver and the evolutionary optimization solver, so that the instruction amplitude of subsequent control cycles can be adaptively compensated.

[0018] Another aspect of this application provides an electric hoist torque adjustment system for handling sudden load changes, the system comprising: The signal acquisition and mutation judgment module is used to collect stator current, rotor speed, drum end torque and lifting end acceleration during the operation of electric hoist. It extracts transient power fluctuation components and calculates the anomaly degree through stator current and rotor speed, and calculates real-time torque change rate and real-time vibration intensity through drum end torque and lifting end acceleration, respectively. The module is weighted and fused to generate a comprehensive confidence score of load mutation. When the comprehensive confidence score of load mutation exceeds the threshold and the speed drop exceeds the tolerance, it is determined that the load mutation condition has been entered. The adaptive scenario tree prediction module is used to construct a multi-branch probability scenario tree for load change conditions and calculate the overall prediction uncertainty. When the time limit is exceeded, the bandpass filter parameters are adjusted and corrected again, and the future torque prediction sequence, the future torque change rate prediction sequence, and the future vibration intensity prediction sequence are output. The multi-constraint working domain generation module is used to obtain the upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity of the high-inertia main drive motor and auxiliary torque compensation device. It compares the future torque change rate prediction sequence and the future vibration intensity prediction sequence with the corresponding upper limit values ​​to generate a change rate constraint mask and a vibration constraint mask. It combines the thermal safety mask calibrated by the basic torque upper limit to form a comprehensive constraint mask to determine the torque working domain. The dual-solver parallel allocation and arbitration module is used to separate the future torque prediction sequence with the highest cumulative probability into low-frequency base load components and high-frequency residual components. The instruction sequence is generated in parallel by a rule-based heuristic solver and an evolutionary optimization solver, and the better one is selected as the feedforward torque instruction through online arbitration. The online feedforward gain correction module is used to collect the actual output torque and compare it with the feedforward torque command during the execution of the feedforward command to obtain the torque deviation sequence. Based on the torque deviation sequence, the feedforward gain coefficient shared by the rule heuristic solver and the evolutionary optimization solver is corrected in real time, so that the command is more in line with the actual dynamic characteristics of the execution unit.

[0019] The beneficial effects of this invention are: This invention overcomes the shortcomings of existing technologies, such as insufficient accuracy in load mutation identification, high prediction uncertainty, lack of parallel fusion of multiple constraints, and inability to balance following error and vibration suppression in torque allocation. By combining weighted fusion of multi-source signals with joint determination of speed drop, the mutation detection delay is compressed to the millisecond level and false alarms are eliminated. Adaptive probability scene trees and closed-loop iteration of filter parameters significantly reduce the uncertainty of future torque prediction, ensuring the reliability of feedforward information under high dynamic conditions. A dynamic torque working domain is generated based on the fusion of three-dimensional masks of rate of change, vibration, and thermal safety, eliminating the risk of execution unit exceeding limits and overheating at the mechanistic level, resulting in a 42% reduction in the torque change rate of the main drive motor. Parallel solution using rule-based heuristics and evolutionary optimization, supplemented by online arbitration, achieves multi-objective collaborative optimization with a 58% reduction in vibration decay time and a 37% reduction in torque following error under strict safety constraints. The recursive correction of the feedforward gain continuously compensates for dynamic differences in the execution unit, forming an adaptive system for all operating conditions. Attached Figure Description

[0020] Figure 1 A schematic diagram of the torque adjustment method for electric hoists to cope with sudden load changes; Figure 2 S100 flowchart of electric hoist torque adjustment method to cope with sudden load changes; Figure 3 S200 flowchart of electric hoist torque adjustment method to cope with sudden load changes; Figure 4 S300 flowchart of electric hoist torque adjustment method to cope with sudden load changes; Figure 5 S400 flowchart of electric hoist torque adjustment method to cope with sudden load changes; Figure 6 S500 flowchart of electric hoist torque adjustment method to cope with sudden load changes. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The core theoretical foundation of this invention is based on transient dynamics of electromechanical systems, stochastic processes and predictive control theory, multi-objective optimization theory, and the thermodynamic derating characteristics of the actuator. By combining torque change rate and vibration intensity characteristics to construct a multi-source fusion load mutation identification criterion, and using historical state transition probabilities and real-time disturbance characteristics to jointly modify and generate a probability scenario tree to characterize the uncertainty of future torque, a three-dimensional mask based on change rate constraints, vibration constraints, and thermal safety constraints characterizes the dynamic capability domain of the actuator. Optimal torque allocation is achieved through parallel solution and online arbitration using rule-based heuristics and evolutionary optimization. Ultimately, a complete method is formed, from mutation identification, multi-path prediction, constraint capability domain generation to cooperative control and adaptive gain correction, realizing highly robust torque coordination and vibration suppression of electric hoists under load mutation conditions.

[0023] The specific embodiments of the present invention will be described in detail below.

[0024] Example 1: Please see Figure 1 It shows an overall schematic diagram of an electric hoist torque adjustment method for dealing with sudden load changes according to an embodiment of the present invention. The method includes: S100: Real-time determination of sudden load changes; S200: Adaptive probabilistic scene tree prediction; S300: Synthetic constraint mask and torque working domain generation; S400: Parallel allocation of dual solvers and online arbitration; S500: Online feedforward gain correction.

[0025] The specific plan is as follows: In an electric hoist torque adjustment method for dealing with sudden load changes, S100 synchronously collects the stator current, rotor speed, drum end torque, and lifting end acceleration of the hoist during operation. It extracts the transient power fluctuation component reflecting the sudden load change and calculates the anomaly degree. At the same time, it differentiates the torque signal to obtain the real-time torque change rate and performs short-time energy integration on the acceleration signal to obtain the real-time vibration intensity. It weights and fuses the three characteristic quantities to generate a comprehensive confidence degree for the sudden load change and combines it with the continuous deviation judgment of the rotor speed drop to confirm the occurrence of the sudden load change condition.

[0026] Please refer to Figure 2 It illustrates a flowchart of an exemplary real-time load change condition determination step S100 of this application, the contents of which include: S110: Synchronous acquisition and preprocessing of multi-source signals; S120: Extraction of transient power fluctuation components and calculation of anomaly degree; S130: Real-time calculation of torque change rate and vibration intensity; S140: Generation of comprehensive confidence scores for load mutations; S150: Logic for determining sudden load changes.

[0027] The S110 synchronously acquires electrical and mechanical signals from the hoisting motor and conditions and aligns the raw signals to provide high-quality time-synchronized data for subsequent feature extraction.

[0028] In this embodiment, a multi-channel synchronous data acquisition module collects four types of signals at a fixed control cycle: stator current, rotor speed, drum end torque, and lifting device end acceleration. The sampling frequency of the acquisition module is set to an integer multiple of the control frequency to ensure the accuracy of subsequent digital signal processing. Each signal passes through an anti-aliasing filter before entering the analog-to-digital conversion. The filter cutoff frequency is determined based on the highest disturbance frequency of the electric hoist drive chain, and a synchronous trigger signal ensures that the sampling time of all channels is strictly consistent.

[0029] In one possible implementation of this embodiment, S110 further includes: S111: The stator current signal is acquired through a Hall current sensor, and the sensor bandwidth needs to cover the main frequency band of power fluctuations; the rotor speed signal is provided by a photoelectric encoder installed on the motor shaft end, and the speed is converted into a digital angular velocity using the M-method; the drum end torque is measured by a strain gauge torque sensor, which is installed between the drum and the reducer output shaft, and the range covers several times the rated torque to cope with impacts; the lifting device end acceleration is acquired by a MEMS acceleration sensor installed on the hook assembly, and only the vertical acceleration component is used in subsequent calculations.

[0030] S112: Before entering the A / D conversion, each analog signal is filtered by a high-order low-pass filter to remove high-frequency noise and potential aliasing components. A Butterworth filter type is selected to ensure passband flatness. The filtered signal is then converted into a digital sequence by a synchronous sampling ADC.

[0031] S113: Due to slight differences in signal link delays between different sensors, a hardware synchronization trigger signal is used to ensure consistent ADC sampling times. The encoder speed signal is then interpolated and resampled in the digital domain to ensure all signals are strictly aligned to the same timestamp grid. The aligned data is then decimated or averaged based on the control cycle to form a discrete sequence synchronized with the control algorithm.

[0032] S120 calculates transient electromagnetic power by aligning stator current and rotor speed signals, then uses a bandpass filter to extract transient power fluctuation components and quantifies the degree of fluctuation anomaly, thereby obtaining the response characteristics of electrical ports to load changes.

[0033] In this embodiment, the instantaneous input power is first calculated based on the stator current and the reconstructed stator voltage, and then combined with the rotor speed information to obtain the transient power signal. Since sudden load changes can induce fluctuations in a specific frequency band in the power signal, the transient power signal is fed into a bandpass filter with adjustable center frequency and bandwidth to obtain the power fluctuation component. The anomaly is defined by comparing the current effective value of the fluctuation component with the historical steady-state reference value.

[0034] In one possible implementation of this embodiment, S120 further includes: S121: In each control cycle, the instantaneous values ​​of the three-phase stator current are acquired, and the stator voltage components are reconstructed according to the voltage equation in the vector control coordinate system, thereby calculating the instantaneous active power. To avoid low-frequency drift introduced by voltage reconstruction error, the instantaneous active power sequence is filtered through a high-pass filter to eliminate steady-state power components, retaining dynamic power changes to obtain the transient power signal.

[0035] S122: The bandpass filter is a second-order Butterworth type, with the initial value of the center frequency set to... The initial bandwidth value is set to . The frequency of the electric hoist drive chain is selected based on its inherent lower limit, ensuring that it falls within the main frequency band of the load change impact energy. This requires covering the spectral width of the shock fluctuations. The filter parameters will be corrected during the adaptive iteration of S200, with the initial values ​​serving as the starting point for the iteration.

[0036] S123: Pass the transient power signal sequence through a bandpass filter to output the fluctuation component sequence. Define the load mutation anomaly degree. This is the normalized ratio of the effective value of the fluctuation component within the sliding window to the effective value of the steady-state component during the reference period. Let the length of the sliding window be... One control cycle, reference steady-state effective value The calculations were performed for the unloaded, uniform-speed lifting phase under the same window length. Valid value. The formula for calculating the degree of abnormality is: . The system performs offline computation and storage during its self-learning phase. When the load is stable... When the value is close to 1, a sudden shock occurs. This increases significantly, thus quantitatively reflecting the degree of abnormality in current power fluctuations.

[0037] The S130 uses the collected torque signal at the drum end and acceleration signal at the spreader end to calculate the real-time torque change rate and real-time vibration intensity, respectively, providing direct mechanical evidence for load change identification and avoiding misjudgments that may occur if only electrical signals are relied upon.

[0038] In this embodiment, the torque change rate is obtained by numerical differentiation of the torque signal, and the vibration intensity is obtained by short-time energy integration of the acceleration signal. Both of these characteristic quantities are closely related to the intensity of load change and have clear physical meaning.

[0039] In one possible implementation of this embodiment, S130 further includes: S131: The drum end torque, after preprocessing in S110, becomes a discrete sequence. The first derivative of the torque is calculated using numerical differentiation. To suppress the amplification effect of high-frequency noise during differentiation, a low-pass filter is applied to the differentiation result. The cutoff frequency is set according to the response bandwidth of the mechanical transmission chain, resulting in a smooth torque change rate sequence, denoted as... .

[0040] S132: The spreader-end acceleration signal contains impact vibration components caused by sudden load changes. The short-time energy integration window width is defined as... , corresponding to Then the sampling point, the first Real-time vibration intensity per control cycle Calculate the sum of squares of the acceleration signals within the window: In the formula, This is the vertical acceleration after removing the gravitational bias. Window width. The selection of the time scale should match the time scale of the impact energy concentration, so as to capture the instantaneous energy release and smooth out random noise.

[0041] S140 will output the anomaly level from S120. Torque variation rate of S130 output and vibration intensity Normalization and weighted fusion are performed to generate a comprehensive confidence score for load mutation. The probability of a sudden load change at the current moment is evaluated from multiple dimensions, including electrical and mechanical ports.

[0042] In this embodiment, to avoid the impact of differences in the dimensions and amplitude ranges of different physical quantities on the fusion effect, the three feature quantities are first mapped to the [0,1] interval, and then the comprehensive confidence level is obtained by weighted summation. The weight allocation is optimized and determined based on the independence and sensitivity of each feature quantity to the abrupt change condition.

[0043] In one possible implementation of this embodiment, S140 further includes: S141: Regarding anomalies A monotonically increasing nonlinear function is used to map the result to the [0,1] interval, causing larger anomalies to rapidly saturate and approach 1. The normalized result is denoted as... For the torque change rate, take the absolute value and divide it by the preset torque change rate threshold. And limit the amplitude to [0,1], to obtain ,in The settings are based on the rated torque and transmission stiffness of the electric hoist. The vibration intensity is also normalized to obtain... , It is several times the average vibration intensity under no-load and normal operating conditions.

[0044] S142: Assign weight coefficients based on the relative importance of the three features. ,satisfy Overall confidence level of load mutation Calculate using the following formula: The weighting coefficients were determined through offline subject working characteristic analysis or expert experience to maximize the mutation detection rate and minimize the false alarm rate.

[0045] Based on the overall confidence level output by S140 and the rotor speed drop, S150 performs a multi-cycle joint judgment to ultimately confirm whether the electric hoist has entered a sudden load change condition. A high overall confidence level alone is not sufficient to confirm a sudden load change condition. Introducing the rotor speed drop as an auxiliary necessary condition can effectively avoid false triggering caused by momentary illusions such as electromagnetic interference.

[0046] In one possible implementation of this embodiment, S150 further includes: S151: Defines the drop in current rotational speed relative to the short-term historical average. ,in For the past The average speed of the rotational speed over one control cycle. The selection of the speed needs to cover the normal speed fluctuation cycle so that the drop during the stable operation phase is close to zero.

[0047] S152: Set the speed drop threshold This threshold is related to the rated speed of the electric hoist and the allowable steady-state fluctuation range. If If so, the current period is marked as the speed deviation period.

[0048] S153: Continuous monitoring And out-of-tolerance marking status. When both of the following conditions are met simultaneously, the electric hoist is determined to have entered a load change condition: Condition one, Exceeding the preset reliability threshold Condition 2, most recent consecutive Each control cycle was marked as a speed over-tolerance cycle. The value of this parameter needs to balance detection speed and anti-interference capability to avoid misjudgment due to speed fluctuations in individual cycles. Once a sudden change condition is detected, the sudden change status flag will be set and maintained for at least the preset duration. This ensures the complete execution of subsequent forecasting and control processes.

[0049] In an electric hoist torque adjustment method for dealing with sudden load changes, S200 constructs a multi-branch probability scenario tree based on historical torque data and real-time disturbance characteristics for the sudden load change conditions confirmed by S150, predicts the future torque trajectory and its derived sequence, and controls the prediction uncertainty by iteratively adjusting the bandpass filter parameters in S120, outputting a prediction sequence with probability weights, providing forward-looking information with uncertainty quantification capabilities for torque allocation.

[0050] Please refer to Figure 3 It illustrates a flowchart of an exemplary adaptive probability scene tree prediction step S200 of this application, the contents of which include: S210: Historical moment trajectory binning statistics and basic transition probability construction; S220: Transition probability correction driven by real-time perturbation features; S230: Multi-branch probabilistic scenario tree construction and future sequence generation; S240: Calculation and evaluation of overall prediction uncertainty for scene tree; S250: Adaptive Filter Parameter Iteration and Scene Tree Reconstruction.

[0051] After determining the load change condition, S210 immediately retrieves the basic load information of the current hoisting task and the historical torque trajectory of the most recent working cycle, performs statistical analysis by working stage, establishes a state transition probability model, and provides data-driven future evolution priors for the scenario tree.

[0052] In this embodiment, the control system of the electric hoist stores the time series of the drum end torque in each working cycle and archives it separately according to the three stages of lifting, translation, and descent. For the historical torque trajectories of the most recent working cycles, the system performs bin-based statistics to construct a discrete state space and a state transition counting matrix.

[0053] In one possible implementation of this embodiment, S210 further includes: S211: Based on the hoisting motor's operating status, contactor signals, or position sensor feedback, determine whether the current moment belongs to the hoisting, translation, or descent phase. Extract all torque trajectories for the corresponding phase from the historical database. Each trajectory is represented as a discrete torque sequence, with the sampling interval being the same as the control cycle.

[0054] S212: Divide the torque range into equal parts There are several state intervals, the number of which is determined based on the torque variation range and accuracy requirements. The torque value at each moment is mapped to its corresponding state interval, forming a torque state label. .

[0055] S213: State pairs of all adjacent moments in the historical trajectory By performing counting, the state transition counting matrix is ​​obtained. ,in Indicates from state Transition to state The total number of times. Basic state transition probability matrix. Depend on The row normalization yields, i.e. If a row is all zeros, it indicates that the state lacks historical data, so it is filled with a uniform distribution to maintain probabilistic integrity.

[0056] S220 utilizes the torque change rate calculated in real time by S100. and vibration intensity The basic transition probability is modified so that the branch probability of the scene tree can reflect the intensity and direction of the current mutation shock, thereby realizing the fusion of prior data and real-time dynamics.

[0057] In this embodiment, the modified model adopts an exponential weighting form, incorporating the normalized value of the real-time disturbance characteristics into the transition probability in a log-linear manner, so that the torque change path consistent with the current impact direction obtains a higher probability.

[0058] In one possible implementation of this embodiment, S220 further includes: S221: Using the same normalization method as S141, obtain the normalized real-time torque change rate at the current moment. and normalized real-time vibration intensity .

[0059] S222: Correct the transition probability calculation. For the transition from the current state... Possible torque state at the next moment after departure Corrected transition probability It is given by the following formula: In the formula, and As an influencing factor, a positive value indicates an increased probability of transitioning to a specific state, and its absolute value is determined through offline parameter sensitivity analysis. As an indicator function, when the state The value is 1 if the direction of the corresponding torque change is consistent with the direction of the current torque change rate; otherwise, it is 0.

[0060] S230 uses the current moment's torque state as the root node, iteratively generates a multi-branch probability scenario tree using the corrected transition probability, and calculates the torque prediction sequence, torque change rate prediction sequence, and vibration intensity prediction sequence corresponding to each future path, providing rich future evolution information for subsequent torque allocation.

[0061] In this embodiment, the depth of the scene tree is the number of prediction steps. Corresponding to the prediction time domain Each node adjusts its transition probability. The branching outwards controls the balance between computational complexity and path coverage integrity by retaining a certain number of child nodes with higher probability.

[0062] In one possible implementation of this embodiment, S230 further includes: S231: The torque state of the root node is taken from the state interval to which the torque belongs at the current moment, and the probability of the root node is set to 1.

[0063] S232: From depth to For each node in the current layer, based on its torque state and corrected transition probability Generate child nodes and assign a branch probability to each child node. The path probability from the root node to a leaf node is the product of the probabilities of all branches on that path. To improve computational efficiency, only a few main branches whose cumulative probability sums reach a preset coverage ratio (e.g., 0.95) can be retained at each level, while low-probability paths are pruned.

[0064] S233: For each complete path in the scene tree, reconstruct the future torque prediction sequence based on the center value of the torque state interval at each step on the path. The torque sequence was subjected to forward time difference using the same numerical differentiation method as S131 to obtain the predicted sequence of future torque change rate. Simultaneously, based on the statistical mapping relationship between the rate of change of torque and vibration intensity in historical data, the corresponding future vibration intensity prediction sequence is calculated. Statistical mapping relationships can be established through linear regression or locally weighted regression and updated periodically.

[0065] S240 calculates the overall prediction uncertainty of the generated scene tree to quantify the dispersion and reliability of the current prediction. When the uncertainty is too high, it indicates that the prediction results are difficult to use for stable control, and S250 needs to be triggered for adaptive adjustment to improve the quality of signal feature extraction and thus reduce prediction uncertainty.

[0066] In this embodiment, the overall prediction uncertainty is measured by the information entropy of the path probability distribution of the leaf nodes in the scene. The larger the information entropy, the more dispersed the possible paths of the future torque, and the lower the confidence of the prediction.

[0067] In one possible implementation of this embodiment, S240 further includes: S241: Extract all arrival depths to the predicted depth The complete path of is denoted by . ,satisfy .

[0068] S242: Overall prediction uncertainty Defined as: Set the maximum allowed limit , The settings are based on the predictive reliability requirements of the electric hoist's operating conditions. If If the prediction result is reliable, the predicted sequence with probability weights will be directly output; if... If the forecast is too uncertain, it indicates that the forecast needs to be adjusted in S250.

[0069] When the prediction uncertainty exceeds the limit, S250 initiates an adaptive iterative loop. By adjusting the center frequency and bandwidth of the bandpass filter in S120, it changes the extraction characteristics of the transient power fluctuation component, thereby recalculating the anomaly degree and subsequent real-time torque change rate and vibration intensity, and further correcting the scene tree branch probability until the uncertainty converges. The maximum number of iterations may be reached below this point. This feedback adjustment mechanism enables closed-loop collaborative optimization of signal feature extraction and prediction models.

[0070] In one possible implementation of this embodiment, S250 further includes: S251: In the first iteration, the center frequency of the bandpass filter is tentatively adjusted. Add a tiny amount and bandwidth Decrease a tiny amount Repeat the entire process from S120 to S240 and record the uncertainty. The change in uncertainty. By comparing the changes in uncertainty under different adjustment directions, the parameter adjustment gradient direction that can reduce uncertainty is determined.

[0071] S252: Update the center frequency and bandwidth along the determined descent direction using a preset step size factor. After the update, re-extract the power fluctuation components, anomaly degree, torque change rate, and vibration intensity, and run S220 to S240 to calculate the new values. After each iteration, check whether the uncertainty meets the requirements.

[0072] S253: When The iteration stops when the condition is met, and the filter parameters at this point are recorded as the optimized configuration for this mutation scenario. The final scene tree and prediction sequence are then output. If the iteration count exceeds a preset limit and convergence is still not achieved, the iteration is forcibly terminated, and a previous iteration is selected. The lowest possible parameter combination is used, while simultaneously outputting uncertainty indicators for upper-level monitoring and alarms.

[0073] Finally, S200 outputs three sets of sequences with probability weights: a set of future torque prediction sequences, a set of future torque change rate prediction sequences, and a set of future vibration intensity prediction sequences. Each path corresponds to a probability weight, providing rich future scenario information for S300 and S400.

[0074] In an electric hoist torque allocation method for dealing with sudden load changes, S300 obtains the configuration, real-time status, and capacity limit of the electric hoist's multi-torque actuators. Using the future torque change rate prediction sequence and future vibration intensity prediction sequence output by S200, combined with the temperature derating relationship, a change rate constraint mask, a vibration constraint mask, and a thermal safety mask are generated. These are then fused into a comprehensive constraint mask and applied to the capacity array to obtain the torque working domain of each actuator at each moment within the prediction period, providing a strict and dynamic safety boundary for torque allocation.

[0075] Please refer to Figure 4 It illustrates a flowchart of an exemplary step S300 for generating a comprehensive constraint mask and torque working domain according to this application, the contents of which include: S310: Execution unit status monitoring and capability limit acquisition; S320: Generation of rate of change constraint mask; S330: Vibration constraint mask generation; S340: Thermal security mask generation; S350: Synthetic constraint mask fusion and torque working domain calculation.

[0076] The S310 collects the current temperature and speed of the main drive motor and auxiliary torque compensation device in real time, and obtains the upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity corresponding to the current state, and establishes the capability baseline of the execution unit.

[0077] In this embodiment, the electric hoist is equipped with at least one high-inertia main drive motor and a fast-response auxiliary torque compensation device. The status parameters of each execution unit are acquired via fieldbus or analog signal channel, and the corresponding upper limit of capability is determined by the equipment specifications and pre-calibrated characteristic curves.

[0078] In one possible implementation of this embodiment, S310 further includes: S311: For the main drive motor, the stator temperature is collected by a temperature sensor embedded in the winding. Rotor speed is obtained through an encoder. For the auxiliary torque compensation device, the temperature of its drive motor is collected. and rotational speed .

[0079] S312: The upper limit of the base torque for each actuator is determined by its speed-torque external characteristic curve. The peak torque curve provided by the manufacturer is stored in the controller in the form of a lookup table or function. By inputting the current speed, the upper limit of the base torque at the current speed can be obtained. ,in This indicates the unit identifier. The external characteristic curve typically represents a constant torque region in the low-speed region and a constant power region in the high-speed region.

[0080] S313: Upper limit of torque change rate The vibration intensity limit is determined by the electrical time constant, mechanical inertia, and drive capability of the actuator; it is a fixed value or a value that varies slightly with the rotational speed, and is calibrated through offline step response testing. These thresholds are set based on the allowable bearing life and installation stiffness standards of the actuator itself, and are also preset in the parameter table. These upper limits constitute the basic constraint boundaries for the dynamic output of the actuator.

[0081] S320 compares the future torque change rate prediction sequence output by S200 with the upper limit of the torque change rate of each unit, generates a change rate constraint mask, and marks the unit whose output is limited at any time in the prediction time domain because the change rate may exceed the limit.

[0082] In this embodiment, since S200 outputs a prediction sequence of multiple paths, the system prioritizes processing the path with the highest cumulative probability, while also taking into account several main paths with probability weights not lower than a certain threshold, in order to form a conservative but not overly constrained mask representation.

[0083] In one possible implementation of this embodiment, S320 further includes: S321: The expected rate of change sequence is obtained by weighting the set of predicted future torque change rates output by S200 according to their respective path probabilities. , Weighted averaging can smooth out extreme values ​​of low-probability paths, making subsequent constraint masks closer to actual probabilities.

[0084] S322: For each execution unit and prediction steps Define the rate of change constraint mask element For: If ,but This indicates that at that moment, the unit's output is limited due to the rate of change constraint; otherwise... This indicates that the rate of change constraint allows for free output. This generates a binary mask matrix with the cell and the prediction time domain as coordinates.

[0085] S330 compares the future vibration intensity prediction sequence output by S200 with the upper limit of vibration intensity of each unit to generate a vibration constraint mask, marking the situation where the output is limited due to excessive vibration, especially protecting the high-frequency response capability of the auxiliary torque compensation device.

[0086] In this embodiment, the vibration intensity prediction sequence is also obtained by path probability weighted averaging. Since vibration intensity mainly affects the high-frequency output of the auxiliary torque compensation device, the mask imposes stricter constraints on the auxiliary torque compensation device.

[0087] In one possible implementation of this embodiment, S330 further includes: S331: For the auxiliary torque compensation device, the additional vibration generated by its high-frequency output may couple with the vibration at the end of the lifting device, therefore an amplification factor is applied. Corrected expected vibration intensity Due to its large inertia, the main drive motor exhibits a negligible vibration amplification effect, resulting in a low amplification factor. .

[0088] S332: If Then the vibration constraint mask element Otherwise, it is 0. This prevents the auxiliary torque compensation device from over-exerting force under high vibration conditions, which could lead to structural resonance or fatigue damage.

[0089] Based on the temperature-torque upper limit derating relationship of the actuator, the S340 marks the time-varying region of the basic torque upper limit as a thermal safety mask to prevent the actuator from being damaged due to overheating and ensure long-term operational reliability.

[0090] In this embodiment, both the main drive motor and the auxiliary torque compensation device are equipped with preset temperature-dependent torque reduction curves, meaning that the maximum allowable output torque gradually decreases as the temperature rises. When the temperature exceeds a certain safety limit, the upper limit of the torque begins to decrease according to a preset rule.

[0091] In one possible implementation of this embodiment, S340 further includes: S341: Considering the short prediction time domain and the slow temperature change of the execution unit, the current sampled temperature is used. As the future The approximate temperature of the step.

[0092] S342: Based on the derating function To obtain the upper limit of the allowable torque at the current temperature. At each moment in the prediction time domain, the upper limit of this thermal torque remains constant, forming a constraint line on the time axis.

[0093] S343: The thermal safety mask does not directly generate 0 / 1 tags, but instead outputs the upper limit of the allowable torque under thermal constraints for each element at each prediction step. To ensure consistency with rate of change constraints and vibration constraints during logical integration, a thermal safety mask element is defined. For: If ,but This indicates that the base upper limit is limited by the thermal derating; otherwise, it is 0. It also stores the actual allowable torque upper limit. .

[0094] S350 performs a logical OR operation on the three masks generated by S320, S330 and S340 element by element according to time and cell, and merges them into a unified comprehensive constraint mask. This mask is then applied to the capability array with cell identifier, prediction time domain and basic torque upper limit as coordinates to obtain the torque working domain that each execution unit is allowed to output at each discrete time in the prediction period, i.e. the safe output range.

[0095] In this embodiment, any limitation in any dimension restricts the upper limit of the output force of the corresponding unit at that moment, reflecting a conservative yet safe design principle. The upper limit of the restricted torque is the most stringent constraint value.

[0096] In one possible implementation of this embodiment, S350 further includes: S351: For each unit and prediction steps Comprehensive constraint mask elements for: ; in This represents the logical "OR" operation. If... This indicates that the unit is subject to at least one constraint at that moment.

[0097] S352: Define Capability Array The initial value is the upper limit of the basic torque. Based on the comprehensive constraint mask and the specific constraint type, determine the upper limit of the constrained torque. If... Then the upper limit of the torque working domain at that moment is .like Then, further determine the dominant constraint: if the rate of change constraint is triggered, reduce the upper limit of torque to the equivalent torque value that meets the upper limit of the rate of change; if the vibration constraint is triggered, reduce it to the maximum allowable torque corresponding to the upper limit of vibration intensity; if the thermal safety constraint is triggered, reduce it to... When multiple constraints are triggered simultaneously, the most stringent upper limit is applied. Ultimately, each execution unit performs its prediction step... Obtain a defined torque working domain This refers to the range of torques allowed to be output at that moment. This working domain sequence will serve as an insurmountable hard constraint for the S400 dual solver.

[0098] In an electric hoist torque distribution method for dealing with sudden load changes, the S400 separates the base load and residual based on the most probable torque trajectory, starts two torque distribution solvers—one rule-based heuristic and the other evolutionary optimization—to solve in parallel, and selects the optimal feedforward torque command through online arbitration to achieve multi-objective coordinated optimization within safety constraints.

[0099] Please refer to Figure 5 It illustrates a flowchart of an exemplary dual-solver parallel assignment and online arbitration step S400 of this application, the contents of which include: S410: Selection of the most likely torque trajectory and separation of load components; S420: Rule-based heuristic solver assignment; S430: Evolutionary Optimizer Solver Search; S440: Online arbitration and feedforward instructions selected.

[0100] S410 selects the future torque prediction sequence with the highest cumulative probability from the scene tree output by S200 as the most likely torque trajectory, and separates the low-frequency base load component and high-frequency residual component through moving average filtering to match the dynamic characteristics of different execution units.

[0101] In one possible implementation of this embodiment, S410 further includes: S411: Traverse all leaf nodes of the scene tree, compare path probabilities, and select the future moment prediction sequence corresponding to the path with the highest probability, denoted as... .

[0102] S412: Apply a length of [length] to the sequence. A moving average filter is used, with the moving average span set according to the time scale of the base charge variation. The filtered result is a low-frequency base charge component sequence. This component changes slowly, reflecting the average load level. The high-frequency residual component is obtained by subtracting the base load component from the original sequence. The residual component contains rapid fluctuations caused by sudden shocks. The base load component is suitable to be handled by a high-inertia main drive motor, while the residual component requires a fast-response auxiliary torque compensation device.

[0103] The S420 operating rule heuristic solver, within the torque operating domain provided by S300, prioritizes the allocation of the base load component to the main drive motor, allocates the high-frequency residual component to the auxiliary torque compensation device, and performs proportional reduction on any over-limit parts, generating a set of deterministic main control command sequences to ensure real-time performance and basic safety.

[0104] In one possible implementation of this embodiment, S420 further includes: S421: For each prediction step ,Will As the initial value of the torque command for the main drive motor ,Will As the initial value of the torque command for the auxiliary torque compensation device .

[0105] S422: Query the torque working domain obtained from S300, and check whether the initial command exceeds the upper limit of the working domain of the corresponding unit. If Exceed Then the main drive motor command will be reduced to The excess portion is added to the auxiliary torque compensation device command. If the auxiliary command also exceeds the limit after addition... Then, both instructions are multiplied by a proportional reduction factor less than 1, ensuring that both remain within their respective operating domains and their sum equals the original total demand torque, maintaining the allocation ratio. The reduction factor is calculated as the maximum common ratio that guarantees neither instruction exceeds its limit and their sum matches. If the original total demand torque itself exceeds the sum of the upper limits of the two unit operating domains, it is proportionally reduced to the sum of the two upper limits. This generates a set of feedforward master control instructions that satisfy all constraints.

[0106] The S430 simultaneously runs an evolutionary optimization solver, aiming to minimize the rate of change of instructions, maximize system efficiency, and minimize the penalty for exceeding limits. It searches for the Pareto front solution set within the same torque operating domain, generating alternative instruction sequences to provide more diverse optimization options for arbitration.

[0107] In one possible implementation of this embodiment, S430 further includes: S431: The main drive motor and auxiliary torque compensation device are in the prediction time domain. The torque command sequence within a step is encoded as a real number vector. The population contains a certain number of individuals. A portion of the individuals in the initial population are randomly generated to ensure that the torque in each step does not exceed the upper limit of the corresponding working domain; another portion of the individuals are generated by applying small perturbations to the solution of S420 to improve the convergence speed and utilize heuristic knowledge.

[0108] S432: An evolutionary optimizer minimizes three objective functions. Objective 1 The instruction change rate penalty, defined as the sum of the squares of the instruction differences between the prediction steps of the two execution units, aims to suppress frequent and drastic adjustments to protect the actuators and reduce energy consumption. Objective Two The system efficiency loss is represented by the degree to which the main drive motor's operating point deviates from the high-efficiency zone and the energy consumption of the auxiliary torque compensation device, and is expressed by a weighted average of the normalized torque square term and the absolute value term. Objective 3 To penalize exceeding limits, a quadratic penalty is applied to any instruction portion that exceeds the upper limit of the working domain, ensuring the enforceability of constraint satisfaction.

[0109] S433: Employs a multi-objective evolutionary algorithm, iteratively evolving within a given computation time limit to obtain a set of Pareto non-dominated solutions. Each solution represents a set of candidate instruction sequences distributed across different trade-off regions in the objective space.

[0110] The S440 uses the online arbitration module to call the real-time collected torque signals at the drum end and acceleration signals at the spreader end from the S100. It performs transient response snapshot evaluations on the main control command sequence output by the rule-based heuristic solver and the alternative command sequence output by the evolutionary optimization solver. It selects the set that is better in terms of both current torque following error and vibration suppression effect as the final feedforward torque command and issues it to each execution unit.

[0111] In one possible implementation of this embodiment, S440 further includes: S441: For each candidate command sequence, using a simplified execution unit response model and the current actual drum end torque and sputtering end acceleration as initial states, the torque following error and vibration intensity response values ​​for the next one or several steps after executing the command are quickly predicted. Since only short-term effects need to be evaluated, the simulation computation is minimal and can be completed online in real time.

[0112] S442: Define Arbitration Indicators It is a weighted sum of the root mean square error of torque following and vibration intensity, and the weighting coefficient can be adjusted to adapt to the different operational requirements for stability and positioning accuracy.

[0113] S443: Select to make The smallest set of command sequences is used as the final feedforward torque command and sent to the main drive motor controller and the auxiliary torque compensation device controller, respectively. When the arbitration indices of the command sequences output by the two solvers are similar, the result of the rule-based heuristic solver is preferred to ensure the interpretability and determinism of the system behavior.

[0114] In an electric hoist torque adjustment method for dealing with sudden load changes, S500 continuously collects the actual output torque of the main drive motor and the auxiliary torque compensation device during the execution of the feedforward command. It compares the torque with the torque of the feedforward torque command selected by S400 point by point to obtain the torque deviation sequence. Based on the deviation sequence, it corrects the feedforward gain coefficient shared by the two solvers in S400 in real time, so that the command is more in line with the actual dynamic characteristics of the execution unit, forming a closed-loop adaptive optimization.

[0115] Please refer to Figure 6 It illustrates a flowchart of an exemplary online feedforward gain correction step S500 of this application, which includes: S510: Actual torque acquisition and deviation sequence generation; S520: Feedforward gain coefficient recursive correction.

[0116] The S510 obtains the actual output torque of the main drive motor and auxiliary torque compensation device through the torque sensor or driver torque estimation function. It compares the torque with the torque of the feedforward torque command to obtain the torque deviation sequence, reflecting the influence of model mismatch and external disturbances.

[0117] In one possible implementation of this embodiment, S510 further includes: S511: The actual electromagnetic torque of the main drive motor is calculated using the torque current component and torque constant output by the driver. The auxiliary torque compensation device is obtained by directly measuring the torque using an independent torque sensor or by estimating the torque through the servo motor current. .

[0118] S512: Calculate torque deviation in each control cycle. and store the length as A sliding window is used to form a bias sequence, which is used to identify the degree of gain mismatch online.

[0119] The S520 uses the bias sequence to correct the feedforward gain coefficient in real time through a recursive least squares method. The gain coefficient acts on the instruction generation model of the two solvers in S400. In the rule-based heuristic solver, the gain coefficient directly scales the base load and residual assignment values. In the evolutionary optimization solver, the gain coefficient affects the initial population generation or the ideal instruction reference in the objective function, enabling future instructions to automatically compensate for the time-varying dynamic characteristics of the execution unit.

[0120] In one possible implementation of this embodiment, S520 further includes: S521: Assuming the ideal gain of the feedforward channel is 1, the actual deviation is mainly caused by gain mismatch, i.e., the actual output torque is approximately... Online estimation is performed using recursive least squares with a forgetting factor. This is to track slow changes in gain.

[0121] S522: To avoid abrupt gain changes due to abnormal deviations, the updated... The gain coefficients are limited to a preset reasonable range, such as a range symmetrical about 1. The corrected gain coefficients are synchronously updated to the shared parameter structure of the two solvers, making the generation of commands in subsequent control cycles more accurate and gradually reducing torque following errors.

[0122] Through the above steps, this embodiment integrates electrical power fluctuations and mechanical dynamic characteristics in load mutation identification, thereby improving the robustness of operating condition determination; the adaptive scene tree enhances the reliability of torque prediction under uncertain environments; the multi-mask fusion working domain ensures the safe output space of multiple execution units under thermal, rate of change, and vibration constraints; the dual solver plus arbitration mechanism balances real-time performance and multi-objective optimization performance; and online gain correction compensates for time-varying characteristics, ultimately achieving stable and efficient torque coordination under sudden shocks.

[0123] Example 2: This embodiment was fully deployed and verified on a certain type of intelligent testing platform for electric hoists. The electric hoist has a rated lifting capacity of 5 tons, the lifting motor is an 11kW variable frequency asynchronous motor, and the rotational inertia of the high-inertia main drive motor (i.e., the lifting motor) is approximately 0.12 kg·m. 2 The auxiliary torque compensation device uses a planetary roller screw mechanism driven by a 2kW servo motor with a response bandwidth of 50Hz, which can quickly generate additional torque to act on the drum drive chain.

[0124] I. Implementation and Configuration; 1. Hardware system configuration: The main controller is an embedded industrial computer running a real-time operating system, with a control cycle of... The signal acquisition uses a synchronous sampling card, which connects to a Hall current sensor, a 1024-line encoder, a strain gauge torque sensor, and a MEMS accelerometer. The auxiliary servo driver communicates via an EtherCAT bus.

[0125] 2. Algorithm parameters: Scene tree state range Predicting step size The corresponding prediction time is 40ms. The bandpass filter has an initial center frequency of 20Hz and an initial bandwidth of 10Hz. The overall confidence threshold is... Number of consecutive out-of-tolerance cycles of speed drop The temperature derating curve is set so that the torque is linearly dated after the stator temperature exceeds 120℃, and 50% remains when the temperature reaches 150℃. The evolutionary optimization population size is 50, and the number of iterations is 30. The initial value of the feedforward gain is 1.0, and the amplitude limit range is [0.8, 1.2].

[0126] II. Experimental conditions; Two typical load change scenarios were designed, each repeated 30 times.

[0127] Condition A – Impact-induced jamming release: A 4.5-ton weight is lifted to a height of 1.5 meters. During the translation process, the weight is momentarily struck against a fixed obstacle to simulate the sudden impact of the weight being jammed and then released. This condition is used to assess the peak suppression capability of the rate of change of torque and vibration intensity.

[0128] Condition B – Rapid Load Change: During the unloaded descent of the spreader, a 2.5-ton weight is suddenly applied at a height of 1 meter, simulating a rapid loading change during loading and unloading. This condition is used to evaluate the system's detection delay and the collaborative response capability of multiple execution units.

[0129] The comparison method is a control strategy that combines fixed proportional torque distribution with traditional threshold detection, while other hardware conditions remain consistent with the method of this invention.

[0130] III. Results and Data; Table 1 shows the key performance indicators of the method of this invention and the comparative method under load abrupt change conditions. The average load abrupt change detection delay of the method of this invention is 8ms, with no false alarms or missed alarms; the detection delay of the comparative method reaches 25ms, and three false alarms occurred under non-abrupt vibration interference. In condition A, the maximum torque change rate of the method of this invention is reduced by 42% compared with the unconstrained condition, while the comparative method is only reduced by 12%. In condition B, the method of this invention reduces the root mean square error of torque following by 37% and shortens the time for the residual vibration intensity at the spreader end to decay to a steady state by 58%.

[0131] Table 1 Comparison of key performance indicators under load change conditions Detailed data on torque distribution and over-limit constraints are shown in Table 2. In 30 tests under operating condition A, the average number of times the upper limit of the main drive motor torque was adjusted due to thermal safety mask triggering was 2.1 times per test, and the average number of times the auxiliary torque compensation device was limited due to vibration constraint mask triggering was 5.7 times per test. The rate of change constraint mask was successfully triggered at the moment of abrupt change, effectively preventing drastic step changes in torque commands. In contrast, the method without a constraint mask mechanism resulted in 8 high-frequency coupling resonance alarms for the auxiliary torque compensation device and 3 shutdowns due to overheat protection for the main drive motor. Through mask fusion, the method of this invention did not experience any overheat protection or torque over-limit alarms for any execution unit in all tests, and the working domain limitation ensured that the output of both execution units remained within the safety boundary.

[0132] Table 2. Torque distribution constraint mask triggering statistics (condition A, 30 experiments) The convergence performance and effect of online feedforward gain correction are shown in Table 3. During the rapid loading process of condition B, the gain coefficient converged to the range of 0.94 to 1.03 within 0.5 seconds, and the torque deviation after convergence was reduced by 65% ​​compared with that before correction. Through online identification using recursive least squares with a forgetting factor, the gain correction process did not cause command oscillation, and the rate of change of the torque command of the auxiliary torque compensation device before and after correction did not increase significantly. This indicates that the online gain correction module can effectively compensate for the time-varying differences in the dynamic characteristics of the execution unit, making the feedforward command more closely match the actual response.

[0133] Table 3. Online correction effect of feedforward gain (condition B, 30 experiments) The experimental data above fully verify the advantages of the method of the present invention in load mutation detection, multi-constraint fusion allocation and online adaptive correction, which can significantly improve the quality of torque cooperative control and vibration suppression effect while ensuring the safety of the execution unit.

[0134] Example 3: An electric hoist torque distribution system for handling sudden load changes, the system comprising: The signal acquisition and mutation judgment module is used to collect stator current, rotor speed, drum end torque and lifting end acceleration during the operation of electric hoist. It extracts transient power fluctuation components and calculates the anomaly degree through stator current and rotor speed, and calculates real-time torque change rate and real-time vibration intensity through drum end torque and lifting end acceleration, respectively. The module is weighted and fused to generate a comprehensive confidence score of load mutation. When the comprehensive confidence score of load mutation exceeds the threshold and the speed drop exceeds the tolerance, it is determined that the load mutation condition has been entered. The adaptive scenario tree prediction module is used to construct a multi-branch probability scenario tree for load change conditions and calculate the overall prediction uncertainty. When the time limit is exceeded, the bandpass filter parameters are adjusted and corrected again, and the future torque prediction sequence, the future torque change rate prediction sequence, and the future vibration intensity prediction sequence are output. The multi-constraint working domain generation module is used to obtain the upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity of the high-inertia main drive motor and auxiliary torque compensation device. It compares the future torque change rate prediction sequence and the future vibration intensity prediction sequence with the corresponding upper limit values ​​to generate a change rate constraint mask and a vibration constraint mask. It combines the thermal safety mask calibrated by the basic torque upper limit to form a comprehensive constraint mask to determine the torque working domain. The dual-solver parallel allocation and arbitration module is used to separate the future torque prediction sequence with the highest cumulative probability into low-frequency base load components and high-frequency residual components. The instruction sequence is generated in parallel by a rule-based heuristic solver and an evolutionary optimization solver, and the better one is selected as the feedforward torque instruction through online arbitration. The online feedforward gain correction module is used to collect the actual output torque and compare it with the feedforward torque command during the execution of the feedforward command to obtain the torque deviation sequence. Based on the torque deviation sequence, the feedforward gain coefficient shared by the rule heuristic solver and the evolutionary optimization solver is corrected in real time, so that the command is more in line with the actual dynamic characteristics of the execution unit.

[0135] Those skilled in the art will understand that embodiments of this application can be embodied in the form of a method, system, or computer program product. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can also take the form of a computer program product embodied on one or more computer-readable storage media containing computer-readable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0136] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make changes and modifications to these embodiments. Therefore, the appended claims are intended to cover the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0137] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. If such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for adjusting the torque of an electric hoist to cope with sudden load changes, characterized in that, include: During the operation of the electric hoist, the stator current, rotor speed, drum end torque and lifting end acceleration are collected. The transient power fluctuation component is extracted by the stator current and rotor speed and the anomaly degree is calculated. The real-time torque change rate and real-time vibration intensity are calculated by the drum end torque and lifting end acceleration, respectively. The load mutation comprehensive confidence degree is generated by weighted fusion. When the load mutation comprehensive confidence degree exceeds the threshold and the speed drop exceeds the tolerance, it is determined that the load mutation condition has been entered. For load change conditions, a multi-branch probability scenario tree is constructed and the overall prediction uncertainty is calculated. When the time limit is exceeded, the bandpass filter parameters are adjusted and corrected again, and the future torque prediction sequence, the future torque change rate prediction sequence, and the future vibration intensity prediction sequence are output. The upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity are obtained for the high-inertia main drive motor and the auxiliary torque compensation device. The future torque change rate prediction sequence and the future vibration intensity prediction sequence are compared with the corresponding upper limit values ​​to generate the change rate constraint mask and the vibration constraint mask. The thermal safety mask calibrated by the upper limit of the basic torque is fused into a comprehensive constraint mask to determine the torque working domain. The future torque prediction sequence with the highest cumulative probability is separated into low-frequency base load components and high-frequency residual components. The instruction sequence is generated in parallel by a rule-based heuristic solver and an evolutionary optimization solver. The better one is selected as the feedforward torque instruction by online arbitration. During the execution of the feedforward instruction, the actual output torque is collected and compared with the feedforward torque instruction to obtain the torque deviation sequence. The feedforward gain coefficient shared by the rule-based heuristic solver and the evolutionary optimization solver is corrected in real time based on the torque deviation sequence, so that the instruction is more in line with the actual dynamic characteristics of the execution unit.

2. The electric hoist torque adjustment method for handling sudden load changes according to claim 1, characterized in that, The extraction of transient power fluctuation components and the calculation of anomaly include: Instantaneous active power is calculated using stator current and reconstructed stator voltage. Steady-state power components are eliminated by a high-pass filter to obtain the transient power signal. The transient power signal is fed into a bandpass filter with adjustable center frequency and bandwidth to extract the power fluctuation component sequence. ; The effective value of the power fluctuation component is calculated using a sliding window and compared with the reference steady-state effective value. Compare and calculate the abnormality of load mutation according to the following formula. : In the formula, The length of the sliding window. These are the effective values ​​of the fluctuation components under the same window length during the no-load steady-state operation phase. This is the sequence number of the current control cycle.

3. The electric hoist torque adjustment method for handling sudden load changes according to claim 2, characterized in that, The calculation of the real-time torque change rate and real-time vibration intensity includes: The first derivative of the torque is calculated by numerical differentiation of the discrete torque sequence at the drum end, and the differentiation result is low-pass filtered to obtain the real-time torque change rate. ; The acceleration signal at the end of the spreader, after removing the gravity bias, is calculated using the following formula. Real-time vibration intensity per control cycle : In the formula, This represents the number of sampling points corresponding to the short-time energy integral window width. This is the vertical acceleration after removing the gravitational bias.

4. The electric hoist torque adjustment method for handling sudden load changes according to claim 3, characterized in that, The method for determining sudden load changes includes: abnormality The real-time torque change rate amplitude and real-time vibration intensity are normalized and mapped to the [0,1] interval to obtain the normalized anomaly degree. Normalized real-time torque change rate and normalized real-time vibration intensity ; According to preset weighting coefficients Perform weighted fusion to generate a comprehensive confidence score for load mutation. : ; Calculate the drop in current rotor speed relative to the short-term historical average. When the drop is continuous The control cycle exceeds the speed drop threshold, and If the preset confidence threshold is exceeded, the electric hoist is determined to have entered a sudden load change condition.

5. The electric hoist torque adjustment method for handling sudden load changes according to claim 4, characterized in that, The multi-branch probability scene tree construction steps include: Historical torque trajectories were statistically analyzed for each of the lifting, translation, and descent phases. The torque value range was discretized into multiple state intervals. The state transition count matrix was statistically analyzed and normalized to obtain the basic state transition probability matrix. ; Using the normalized real-time torque change rate and normalized real-time vibration intensity The base transition probability is corrected to obtain the corrected transition probability. : In the formula, This represents the current torque state. This represents the possible torque state at the next moment. and As a preset impact factor, The indicator function represents the possible torque state at the next moment. The value is 1 if the direction of the corresponding torque change is consistent with the direction of the real-time torque change rate; otherwise, it is 0. Using the current torque state as the root node, the tree expands outward layer by layer according to the corrected transition probability to the preset prediction steps, thus constructing a multi-branch probability scenario tree.

6. The electric hoist torque adjustment method for coping with sudden load changes according to claim 5, characterized in that, The multi-branch probability scene tree construction step also includes: Reconstruct the future torque prediction sequence based on the center value of the torque state interval at each step on each path in the scene tree; By performing forward time difference on the future torque prediction sequence, the future torque change rate prediction sequence is obtained; Based on the statistical mapping relationship between the rate of change of torque and vibration intensity, the future vibration intensity prediction sequence is calculated. The probability of each path is output as the probability weight of the corresponding prediction sequence.

7. The electric hoist torque adjustment method for handling sudden load changes according to claim 1, characterized in that, The calculation of overall prediction uncertainty and the correction of bandpass filter parameters after exceeding the time limit include: Collect the path probabilities of all complete paths reaching the predicted depth in the scene tree. Calculate the overall prediction uncertainty using the following formula. : ; when Exceeding the allowed limit At that time, along with The center frequency and bandwidth of the bandpass filter are iteratively adjusted in the descent gradient direction. The transient power fluctuation components, anomalies, real-time torque change rate and real-time vibration intensity are re-extracted, and the scenario tree branch probability is corrected until the overall prediction uncertainty converges to below the allowable upper limit or reaches the maximum number of iterations.

8. The electric hoist torque adjustment method for handling sudden load changes according to claim 1, characterized in that, The steps for generating the rate of change constraint mask, vibration constraint mask, and thermal safety mask include: The absolute value of the predicted future torque change rate sequence is taken by weighting the path probability and comparing it with the upper limit of the torque change rate of each execution unit to generate a change rate constraint mask. The predicted sequence of future vibration intensity is weighted by path probability, and a preset vibration sensitivity amplification factor is applied to the auxiliary torque compensation device. The result is then compared with the upper limit value of vibration intensity for each execution unit to generate a vibration constraint mask. Based on the preset temperature-torque upper limit derating relationship, the allowable torque upper limit corresponding to the currently collected execution unit temperature is compared with the basic torque upper limit to calibrate the thermal safety mask.

9. The electric hoist torque adjustment method for coping with sudden load changes according to claim 8, characterized in that, The steps for determining the torque operating domain using the comprehensive constraint mask include: Rate of change constraint mask Vibration constraint mask thermal security mask Perform a logical OR operation element-wise on the prediction step and each cell to obtain the synthesized constraint mask. : In the formula Indicates the execution unit identifier. Indicates the prediction step. This represents the logical "OR" operator. when When the indication is limited, the corresponding unit in the prediction step is determined according to the dominant constraint type. The upper limit of the torque is reduced to a value that satisfies the corresponding constraints, forming the torque working domain. .

10. The electric hoist torque adjustment method for handling sudden load changes according to claim 1, characterized in that, The instruction sequence generation step includes: The future torque prediction sequence with the highest cumulative probability is smoothed by using a moving average filter to separate the low-frequency base charge component and the high-frequency residual component. The operating rule-based heuristic solver prioritizes the allocation of low-frequency base load components to the high-inertia main drive motor within the torque operating domain, allocates high-frequency residual components to the auxiliary torque compensation device, and performs proportional reduction on any out-of-limit parts to generate a main control command sequence. Simultaneously, an evolutionary optimization solver is run to search for Pareto front solutions within the same torque operating domain, aiming to minimize the rate of change of instructions, maximize system efficiency, and minimize over-limit penalties, thereby generating alternative instruction sequences.

11. The electric hoist torque adjustment method for coping with sudden load changes according to claim 10, characterized in that, The step of selecting the feedforward torque command includes: The system calls upon the real-time collected torque signals at the drum end and acceleration signals at the spreader end to perform transient response prediction on the main control command sequence and alternative command sequences, and calculates the weighted arbitration index of torque following error and vibration intensity response. The set of instructions that minimizes the weighted arbitration index is selected as the final feedforward torque command and sent to the high-inertia main drive motor and auxiliary torque compensation device.

12. The electric hoist torque adjustment method for handling sudden load changes according to claim 1, characterized in that, The torque deviation sequence is obtained through the following methods: The actual electromagnetic torque of the high-inertia main drive motor is calculated by the torque current and torque constant output by the driver, and the actual output torque of the auxiliary torque compensation device is estimated by the torque sensor or servo motor current. In each control cycle, the difference between the feedforward torque command and the actual output torque is calculated point by point to obtain the torque deviation sequence and store it in the sliding window.

13. The electric hoist torque adjustment method for coping with sudden load changes according to claim 12, characterized in that, The feedforward gain coefficient shared by the real-time correction rule heuristic solver and the evolutionary optimization solver based on the torque deviation sequence includes: A feedforward channel gain mismatch model is established, and the actual output torque is expressed as the product of the feedforward gain coefficient and the feedforward torque command. The recursive least squares method with a forgetting factor is used to estimate the feedforward gain coefficient online using the torque deviation sequence; After limiting the estimated value to a preset range, it is synchronously updated to the instruction generation model of the rule-based heuristic solver and the evolutionary optimization solver, so that the instruction amplitude of subsequent control cycles can be adaptively compensated.

14. An electric hoist torque adjustment system for coping with sudden load changes, characterized in that, Includes a module for performing the method as described in any one of claims 1-13: The signal acquisition and mutation judgment module is used to collect stator current, rotor speed, drum end torque and lifting end acceleration during the operation of electric hoist. It extracts transient power fluctuation components and calculates the anomaly degree through stator current and rotor speed, and calculates real-time torque change rate and real-time vibration intensity through drum end torque and lifting end acceleration, respectively. The module is weighted and fused to generate a comprehensive confidence score of load mutation. When the comprehensive confidence score of load mutation exceeds the threshold and the speed drop exceeds the tolerance, it is determined that the load mutation condition has been entered. The adaptive scenario tree prediction module is used to construct a multi-branch probability scenario tree for load change conditions and calculate the overall prediction uncertainty. When the time limit is exceeded, the bandpass filter parameters are adjusted and corrected again, and the future torque prediction sequence, the future torque change rate prediction sequence, and the future vibration intensity prediction sequence are output. The multi-constraint working domain generation module is used to obtain the upper limit of the basic torque, the upper limit of the torque change rate, and the upper limit of the vibration intensity of the high-inertia main drive motor and auxiliary torque compensation device. It compares the future torque change rate prediction sequence and the future vibration intensity prediction sequence with the corresponding upper limit values ​​to generate a change rate constraint mask and a vibration constraint mask. It combines the thermal safety mask calibrated by the basic torque upper limit to form a comprehensive constraint mask to determine the torque working domain. The dual-solver parallel allocation and arbitration module is used to separate the future torque prediction sequence with the highest cumulative probability into low-frequency base load components and high-frequency residual components. The instruction sequence is generated in parallel by a rule-based heuristic solver and an evolutionary optimization solver, and the better one is selected as the feedforward torque instruction through online arbitration. The online feedforward gain correction module is used to collect the actual output torque and compare it with the feedforward torque command during the execution of the feedforward command to obtain the torque deviation sequence. Based on the torque deviation sequence, the feedforward gain coefficient shared by the rule heuristic solver and the evolutionary optimization solver is corrected in real time, so that the command is more in line with the actual dynamic characteristics of the execution unit.

Images