Marine crane group dynamic power prediction and cooperative control system and method

Abstract

The invention discloses a dynamic power prediction and cooperative control system and method for an ocean crane group, and belongs to the technical field of ocean engineering equipment control. According to the method, encoder position information, load information, operation instructions and motor currents are collected from all cranes at a fixed sampling period, and multi-dimensional feature vectors are constructed after filtering and normalization processing; adopting a Kalman filtering algorithm to identify the actual load mass online and correct a power prediction value; calculating a priority index, and carrying out multi-machine power collaborative distribution under the power constraint of the generator; the position of each crane boom end is predicted based on kinematics extrapolation, and a predicted distance is calculated to generate a coordinated deceleration instruction; and issuing the instruction to each crane control node for execution through a distributed soft bus based on an OpenHarmony operating system. According to the method, power dynamic prediction, collaborative optimization distribution and interference active pre-judgment are realized, and the power utilization efficiency and the safety are improved when a plurality of cranes work at the same time.

Description

Technical Field

[0001] This invention belongs to the field of marine engineering equipment control technology, specifically relating to a dynamic power prediction and collaborative control system and method for marine crane groups. Background Technology

[0002] Offshore cranes are crucial loading and unloading equipment on bulk carriers and container ships, used for cargo lifting and transfer operations. In practice, ships are typically equipped with multiple cranes to improve loading and unloading efficiency. However, the output power of ship generators is limited. When multiple cranes operate simultaneously, the sum of their power demands may exceed the generator's supply capacity, necessitating coordinated power allocation among the cranes. Furthermore, the working areas of multiple cranes may overlap, posing a risk of collisions during simultaneous operations, requiring monitoring and control of spatial interference between cranes.

[0003] Existing marine crane control systems mostly employ static allocation strategies for power management, meaning they pre-set power limits for each crane and do not dynamically adjust based on actual load and operating conditions during operation. Since crane power demand fluctuates with load changes, static allocation strategies struggle to accurately match actual power requirements, leading to low generator power utilization or insufficient power when multiple cranes are simultaneously operating under heavy loads. Furthermore, current technologies lack accurate methods for predicting crane power demand, making it impossible to anticipate power change trends and resulting in delayed power allocation decisions.

[0004] Existing crane collision avoidance systems mostly employ a passive zone restriction method, which involves pre-defining the working area boundaries for each crane and triggering a stop protection mechanism when the crane moves near the boundary. This passive method cannot predict future interference risks based on the crane's movement status, often triggering a stop only when the crane is already very close to the collision point, leading to frequent emergency stops. Emergency stops cause significant impacts on the crane's mechanical structure and the load, affecting equipment lifespan and operational safety. Furthermore, passive methods require pre-defined fixed working areas, lacking flexibility and making it difficult to adapt to complex and changing operational scenarios.

[0005] Existing marine crane control systems mostly adopt a centralized architecture and foreign industrial control software, which limits system scalability and poses potential risks regarding software autonomy and controllability. With the development of the localization and intelligentization of marine engineering equipment, there is an urgent need to develop a distributed crane group control system based on a domestic operating system to achieve intelligent functions such as dynamic power prediction, multi-machine collaborative allocation, and active interference prediction, thereby improving the operating efficiency and safety of marine cranes. Summary of the Invention

[0006] To address the problems existing in the background art, the present invention provides a method for dynamic power prediction and coordinated control of marine crane groups, comprising the following steps: S1. Data Acquisition and Preprocessing: At a fixed sampling period, the lifting position, luffing angle, and slewing angle output by the encoders of each crane, the hook load value output by the load sensor, the operation command value output by the master handle, and the motor current value fed back by the driver are collected; the collected raw data are filtered to remove noise, and then normalized to map various types of data to a unified range, and a multi-dimensional feature vector containing historical power sequences, operation commands, and load information is constructed. S2. Dynamic Power Prediction: Input the multi-dimensional feature vector constructed in S1 into the LSTM neural network, and calculate the hidden state sequence of each time step through the LSTM unit; calculate the attention weight of each time step based on the hidden state sequence, and sum the attention weight with the hidden state to obtain the context vector; input the context vector into the output layer to output the power prediction sequence of each crane for multiple future time steps; S3. Load Online Identification: Establish a state-space model including position, velocity, and load mass. Use the Kalman filter algorithm to predict and update the state using encoder position observations. Extract the load mass estimate from the updated state vector. Correct the power prediction value output by S2 based on the ratio of the identified load mass to the reference load mass. S4. Multi-machine power collaborative allocation: Calculate the power allocation priority index based on the current load mass, lifting height and operation urgency of each crane; take the minimum power fluctuation and the minimum priority weighted deviation as the optimization objectives, and take the total generator power constraint, single machine power constraint and power change rate constraint as boundary conditions, and use the rolling time domain optimization method to solve the optimal power allocation value of each crane, and output the optimal value of the current time step as the power allocation command; S5. Multi-machine interference prediction and coordinated control: Obtain the current amplitude angle, slewing angle and boom length of each crane, and calculate the angular velocity based on the angle difference between adjacent sampling periods; predict the future spatial coordinates of each crane boom end based on the current pose and angular velocity using kinematic formulas; calculate the predicted distance between the boom ends of any two cranes, calculate the deceleration coefficient based on the relationship between the predicted distance and the warning threshold and the stopping threshold, and multiply the deceleration coefficient by the reference speed to generate a coordinated deceleration command; S6. Distributed Communication and Execution: Power allocation instructions and coordinated deceleration instructions are encapsulated into data frames and sent to each crane control node via a soft bus; after parsing the data frames, each control node controls the hydraulic valve to adjust the hydraulic system output according to the power allocation instructions, and controls the driver to adjust the motor speed according to the coordinated deceleration instructions.

[0007] Furthermore, S2 specifically includes: S21. Input the multidimensional feature vector into each LSTM unit of the LSTM network in sequence. Each LSTM unit outputs the hidden state at the corresponding time step to obtain the hidden state sequence. S22. Calculate the attention weights at each time step using the following formula: ; in, For the first Attention weights for each time step; It is an exponential function; For the first The attention score at each time step is calculated from the hidden state through a linear transformation and activation function; Sum of the attention score index values ​​for all time steps; S23. The context vector is obtained by weighted summing of the attention weights and the hidden state, calculated using the following formula: ; in, For context vectors; This represents the summation over all time steps; For the first Attention weights for each time step; For the first The hidden state vector at each time step; S24. Input the context vector into the fully connected output layer to obtain the power prediction sequence, calculated using the following formula: ; in, For power prediction sequence (kW); This is the output layer weight matrix; For context vectors; This is the output layer bias vector.

[0008] Furthermore, the load quality identification in S3 specifically includes: S31. Define the state vector ,in, for Time-state vector; for Momentary lifting position (m); for Lifting speed at any time (m / s); for Estimated load mass at any time (kg); S32. According to Time-state estimation prediction The state at time step is predicted using the following formula: ; in, for Predicted state value at any given time; This is the state transition matrix; for State estimate at time; To control the input matrix; for Constant driving torque (N·m); S33. Update the state estimate using encoder position observations. The update formula is: ; in, for State estimate at time; This is the predicted state value; The Kalman gain matrix; for Time and location observation (m); The observation matrix; The third component is the identified load quality. ; S34. Correct the power prediction value based on the identified load. The correction formula is as follows: ; in, The corrected power prediction value (kW); The original power prediction value (kW); To identify the load mass (kg); Reference load mass (kg).

[0009] Furthermore, the power cooperative allocation in S4 specifically includes: S41. Calculate the priority index of each crane. The calculation formula is as follows: ; in, For the first Crane priority index; This is the load weighting coefficient; For the first Current load capacity of the crane (kg); Rated maximum load mass (kg); This is a highly weighted coefficient; For the first Current lifting height of the crane (m); The safe height threshold (m); This is the urgency weighting coefficient; For the first Emergency response level for operating a multi-stage crane; , , The sum is 1; S42. Construct the optimization objective function: ; in, The objective function value; To sum over each time step in the prediction time domain; To sum over all cranes; Weights for power change penalties; For the first The first crane Step power change (kW); Weights for bias penalty; Priority index; For the first The first crane Step reference power (kW); For the first The first crane Step-by-step power distribution (kW); S43. Set constraints, the total power constraint is: ; in, For all cranes Sum of step power (kW); The generator's available power (kW); Reserved power (kW); single unit power constraint is: ; in, This represents the minimum power output of a single unit (kW). This represents the maximum power of a single unit (kW). S44. The objective function is minimized under constraints using a quadratic programming algorithm to obtain the optimal power allocation sequence. The value of the first time step is taken as the current power allocation command output.

[0010] Furthermore, the interference prediction and coordinated control in S5 specifically includes: S51. Calculate the angular velocity of each crane. The formula for calculating the luffing angular velocity is: ; in, For the first Lamp-span angular velocity of the crane (rad / s); for Time-varying angle (rad); for Time-varying angle (rad); Sampling period (s); angular velocity Calculate using the same method; S52. Predict the spatial coordinates of the arm tip. The prediction formula is as follows: ; in, , , For the first Predicted coordinates (m) of the boom end of the gantry crane; The arm length is (m). The current amplitude angle (rad); The amplitude angular velocity is (rad / s). For the predicted time (s); The current rotation angle (rad); Angular velocity (rad / s); The height of the base is (m). It is a cosine function; It is a sine function; S53. Calculate the predicted distance between the boom ends of the two cranes. The calculation formula is as follows: ; in, For cranes With crane Predicted distance at the end of the arm (m); , , For cranes Predicted coordinates of the arm tip (m); , , For cranes Predicted coordinates of the arm tip (m); S54. Calculate the deceleration coefficient based on the predicted distance: when hour, ;when hour, ;when hour, ;in, This is the deceleration coefficient; The predicted distance (m); The warning distance threshold (m); The parking distance threshold (m); S55. Generate a coordinated deceleration command; the calculation formula is as follows: ; in, For speed commands; For reference speed; This is the deceleration coefficient.

[0011] This invention also provides a dynamic power prediction and coordinated control system for marine crane fleets, including: The main controller, deployed in the ship's central control room, includes a power prediction module, a load identification module, a cooperative allocation module, an interference prediction module, and a communication management module. The power prediction module performs dynamic power prediction in step S2. The load identification module, data-connected to the power prediction module, performs load identification in step S3 and outputs correction coefficients to the power prediction module. The cooperative allocation module, data-connected to the power prediction module, performs cooperative power allocation in step S4. The interference prediction module performs interference prediction and coordinated control in step S5. The communication management module, data-connected to both the cooperative allocation module and the interference prediction module, performs data frame encapsulation and command issuance in step S6. Multiple crane control nodes are provided, each corresponding to one crane. Each crane control node is bidirectionally connected to the communication management module of the main controller via a soft bus. Each crane control node is equipped with a data acquisition unit and an execution control unit. The data acquisition unit is used to perform data acquisition and preprocessing in S1. The execution control unit is connected to the data acquisition unit and is used to perform instruction parsing and motion control in S6. Encoders: Each crane is equipped with multiple encoders, which are installed on the hoisting mechanism, luffing mechanism and slewing mechanism respectively. The signal output terminal of each encoder is connected to the position signal input terminal of the data acquisition unit of the corresponding crane control node. Each crane is equipped with one load sensor, which is installed at the hook assembly. The signal output terminal of the load sensor is connected to the load signal input terminal of the data acquisition unit of the corresponding crane control node. Each crane is equipped with a master control handle, which is installed in the operator's cab. The signal output terminal of the master control handle is connected to the command signal input terminal of the data acquisition unit of the corresponding crane control node. Hydraulic valves are provided in each crane, and the control signal input terminal of the hydraulic valve is connected to the hydraulic control output terminal of the execution control unit of the corresponding crane control node. Each crane is equipped with a driver. The control signal input terminal of the driver is connected to the drive control output terminal of the execution control unit of the corresponding crane control node, and the current feedback output terminal of the driver is connected to the current signal input terminal of the data acquisition unit of the corresponding crane control node.

[0012] In a preferred embodiment, the power prediction module includes an LSTM network unit and an attention calculation unit; the input of the LSTM network unit is connected to the data acquisition unit via a communication management module to receive multi-dimensional feature vectors and output a hidden state sequence; the input of the attention calculation unit is connected to the output of the LSTM network unit to calculate attention weights and output a context vector; the output of the attention calculation unit is connected to the output layer, and the output layer outputs a power prediction sequence.

[0013] In a preferred embodiment, the collaborative allocation module includes a priority calculation unit, an objective function construction unit, and an optimization solution unit. The input of the priority calculation unit is connected to the data acquisition unit via a communication management module to receive the load and position information of each crane and output a priority index. The input of the objective function construction unit is connected to the output of both the priority calculation unit and the power prediction module. The input of the optimization solution unit is connected to the output of the objective function construction unit, and the optimization solution unit also communicates with the ship power plant management system to obtain the available power of the generator as a constraint.

[0014] In a preferred embodiment, the interference prediction module includes a velocity calculation unit, a position prediction unit, and a deceleration command generation unit. The input end of the velocity calculation unit is connected to the data acquisition unit via a communication management module to receive encoder data from each crane and output angular velocity. The input end of the position prediction unit is connected to the output end of the velocity calculation unit to output predicted coordinates of each crane boom end. The input end of the deceleration command generation unit is connected to the output end of the position prediction unit to calculate the predicted distance and deceleration coefficient and output a coordinated deceleration command.

[0015] In the preferred embodiment, the main controller and each crane control node are industrial control computers based on the OpenHarmony operating system; the soft bus adopts a distributed soft bus protocol, with the main controller as the master node and each crane control node as the slave node, and the master node and slave nodes exchange data in a publish-subscribe mode; the communication management module is also configured with a Modbus protocol interface for communication connection with the ship power plant management system.

[0016] The beneficial effects achieved by this invention are as follows: This invention employs an LSTM neural network combined with an attention mechanism to dynamically predict the future power demand of each crane in the time domain. The attention mechanism automatically learns the importance of each time step in the historical power sequence, allowing the prediction model to selectively focus on historical information more relevant to the current prediction task. Compared to traditional static power estimation methods, this dynamic prediction method can anticipate the power demand trends of each crane, providing accurate input data for subsequent coordinated power allocation. Furthermore, this invention uses an adaptive Kalman filter algorithm to identify the actual load quality of each crane in real time and corrects the power prediction values ​​based on the identification results. This allows the power prediction model to adapt to different load conditions, improving the generalization ability and accuracy of the prediction.

[0017] This invention employs a rolling time-domain optimization method based on model predictive control for multi-crane power collaborative allocation. An objective function is constructed, incorporating power variation penalties and priority-weighted deviation penalties. Under constraints of total generator power and individual crane power, the optimal power allocation value for each crane is solved, achieving global coordinated optimization of the power demands of multiple cranes. The priority index designed in this invention comprehensively considers factors such as load quality, lifting height, and operational urgency, ensuring that high-priority cranes receive power allocations closer to their needs. When generator power is limited, priority is given to tasks with higher safety requirements. The rolling time-domain optimization strategy enables the system to dynamically adjust the power allocation scheme based on real-time status information, providing a rapid response capability to environmental changes and disturbances.

[0018] This invention employs a kinematic extrapolation-based method to predict the future spatial position of each crane boom. By calculating the predicted distance between any two cranes and comparing it with a safety threshold, it achieves early prediction of the risk of spatial interference between cranes. The piecewise linear deceleration function designed in this invention calculates the deceleration coefficient based on the relationship between the predicted distance and the warning and stopping thresholds, enabling the cranes to decelerate smoothly and proportionally when entering the warning area. This avoids the sudden stop problem in traditional passive area restriction methods, reduces the impact on the mechanical structure and the hoisted goods, extends the service life of the equipment, and improves operational safety.

[0019] This invention employs a distributed system architecture, where the main controller and each crane control node communicate via a distributed soft bus based on the OpenHarmony operating system, supporting flexible configuration and expansion of the equipment. Utilizing a domestically developed and controllable operating system and communication protocol eliminates reliance on foreign industrial control software, thereby improving the autonomy of marine engineering equipment control systems. The communication management module is also equipped with a Modbus protocol interface, enabling easy integration with existing ship power plant management systems to obtain generator operating status information and achieve coordinated operation between the crane control system and the ship's power system. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for dynamic power prediction and coordinated control of marine crane fleets. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a dynamic power prediction and collaborative control system for marine crane fleets, which adopts a distributed architecture and consists of a main controller, multiple crane control nodes, sensors, and actuators.

[0023] The main controller, located in the ship's central control room, serves as the computational and decision-making center for the entire system. It contains a power prediction module, a load identification module, a cooperative allocation module, an interference prediction module, and a communication management module. The power prediction module performs dynamic power prediction in step S2, receiving multi-dimensional feature vectors from each crane, running an LSTM neural network and attention mechanism algorithm, and outputting a power prediction sequence for each crane. The load identification module, data-connected to the power prediction module, performs load identification in step S3, running a Kalman filter algorithm, outputting a load quality estimate, and passing a correction coefficient to the power prediction module for power prediction value correction. The cooperative allocation module, also data-connected to the power prediction module, performs power cooperative allocation in step S4, receiving the power prediction sequence and priority parameters, running a model predictive control algorithm, and outputting power allocation commands for each crane. The interference prediction module performs interference prediction and coordinated control in step S5, receiving pose and velocity data from each crane, running kinematic extrapolation and distance calculation algorithms, and outputting coordinated deceleration commands. The communication management module is connected to the collaborative allocation module and the interference prediction module respectively, and is used to perform data frame encapsulation and instruction issuance in step S6, and manage communication with each crane control node.

[0024] Multiple crane control nodes, each corresponding to a specific crane, are deployed within the electrical control cabinet of each crane. Each crane control node is bidirectionally connected to the main controller's communication management module via a soft bus. Each crane control node contains a data acquisition unit and an execution control unit. The data acquisition unit performs data acquisition and preprocessing in step S1, connecting to various sensors to collect data at a fixed sampling period, performing filtering and normalization, constructing a multi-dimensional feature vector, and uploading it to the main controller via the communication management module. The execution control unit is data-connected to the data acquisition unit and performs instruction parsing and motion control in step S6, receiving power distribution instructions and coordinated deceleration instructions from the main controller, and generating hydraulic valve control signals and actuator control signals.

[0025] Encoders are used to detect the position information of various mechanisms of the crane. Each crane is equipped with multiple encoders, which are installed on the drive shafts of the hoisting mechanism, luffing mechanism, and slewing mechanism, respectively. The signal output terminal of each encoder is connected to the position signal input terminal of the data acquisition unit of the corresponding crane control node. Absolute encoders are preferred, as they can directly obtain the absolute position after the system is powered on without the need for zeroing operations.

[0026] Load sensors are used to detect the weight of the hoisted goods. Each crane is equipped with one load sensor, installed at the connection between the hook assembly and the wire rope. The signal output terminal of the load sensor is connected to the load signal input terminal of the data acquisition unit of the corresponding crane control node. Strain gauge load sensors are preferred due to their high measurement accuracy and fast response speed.

[0027] The master control handle is used to input operator commands. Each crane is equipped with one master control handle, which is installed on the control panel in the operator's cab. The signal output terminal of the master control handle is connected to the command signal input terminal of the data acquisition unit of the corresponding crane control node. The master control handle is preferably a proportional type, which outputs an analog voltage signal proportional to the handle deflection angle, enabling stepless speed control.

[0028] Hydraulic valves are used to control the output of the hydraulic system. Each crane is equipped with a hydraulic valve assembly, including multiple hydraulic valves that control hoisting, luffing, and slewing movements. The control signal input terminals of the hydraulic valves are connected to the hydraulic control output terminals of the corresponding crane control node's execution control unit. By adjusting the opening degree of the hydraulic valves, the flow rate and pressure of the hydraulic oil can be controlled, thereby controlling the output power of the hydraulic motor or hydraulic cylinder.

[0029] The driver is used to control the operation of the motor. Each crane is equipped with a driver. The driver's control signal input is connected to the drive control output of the execution control unit of the corresponding crane control node, and the driver's current feedback output is connected to the current signal input of the data acquisition unit of the corresponding crane control node. A variable frequency drive is preferably used, which achieves stepless adjustment of the motor speed by regulating the output frequency. The current feedback signal is used for calculations by the load identification module.

[0030] The power prediction module's internal structure includes an LSTM network unit and an attention calculation unit. The input of the LSTM network unit is connected to the data acquisition unit via a communication management module, receiving multi-dimensional feature vectors and running the LSTM network algorithm to output a hidden state sequence. The input of the attention calculation unit is connected to the output of the LSTM network unit, calculating attention weights and generating a context vector. The output of the attention calculation unit is connected to the output layer, which maps the context vector to a power prediction sequence.

[0031] The collaborative allocation module's internal structure includes a priority calculation unit, an objective function construction unit, and an optimization solution unit. The input of the priority calculation unit is connected to the data acquisition unit via a communication management module, used to receive load mass and lifting height information from each crane and calculate the priority index for each crane. The input of the objective function construction unit is connected to the outputs of both the priority calculation unit and the power prediction module, used to construct an optimization objective function based on the priority index and power prediction values. The input of the optimization solution unit is connected to the output of the objective function construction unit, used to solve for the minimum value of the objective function under constraints. The optimization solution unit also communicates with the ship's power plant management system to obtain the available generator power in real time as a total power constraint.

[0032] The internal structure of the interference prediction module includes a velocity calculation unit, a position prediction unit, and a deceleration command generation unit. The input of the velocity calculation unit is connected to the data acquisition unit via a communication management module, used to receive encoder angle data from each crane and calculate the luffing angular velocity and slewing angular velocity of each crane. The input of the position prediction unit is connected to the output of the velocity calculation unit, used to perform kinematic extrapolation based on the current pose and angular velocity, and output the predicted coordinates of the boom ends of each crane. The input of the deceleration command generation unit is connected to the output of the position prediction unit, used to calculate the predicted distance between each pair of cranes, calculate the deceleration coefficient based on the relationship between distance and a threshold, and generate a coordinated deceleration command.

[0033] The main controller and each crane control node all use industrial control computers based on the OpenHarmony operating system. OpenHarmony is a distributed operating system for all scenarios, supporting various device types and featuring distributed soft bus and distributed data management. Using a unified operating system platform simplifies system development and maintenance, enabling independent software control. The soft bus adopts a distributed soft bus protocol, with the main controller acting as the master node and each crane control node as a slave node. Data exchange between the master and slave nodes uses a publish-subscribe model, supporting one-to-many message broadcasting and point-to-point data transmission. The communication management module is also configured with a Modbus protocol interface for communication with the ship's power plant management system to obtain generator operating status and available power information. Modbus is a widely used communication protocol in the industrial field, offering good compatibility and easy integration with existing ship electrical equipment.

[0034] Reference Figure 1 This invention also provides a method for dynamic power prediction and coordinated control of marine crane groups. This method is applicable to scenarios where multiple cranes are configured on bulk carriers or container ships, and is particularly suitable for situations where multiple cranes share power from a ship's generator and their working areas overlap. In practical applications, ships typically have four cranes. The power demand of each crane fluctuates with load changes during operation, while the output power of the ship's generator is limited. Therefore, it is necessary to coordinate the power allocation among multiple cranes, and simultaneously predict and control spatial interference between cranes. The method of this invention includes the following steps.

[0035] Step S1 involves data acquisition and preprocessing. In this step, the system collects various sensor data from each crane at a fixed sampling period. The selection of the sampling period needs to comprehensively consider control accuracy and computational load; the preferred sampling period is 50ms to 200ms. The acquired data includes the lifting position, luffing angle, and slewing angle output by the encoder. This positional information is used to determine the crane's current spatial attitude. Simultaneously, the hook load value output by the load sensor is acquired, reflecting the actual weight of the currently lifted cargo. Furthermore, the operation command value output by the master control handle is acquired, representing the operator's intention and operational range. The motor current value fed back by the driver also needs to be acquired; this current value is related to the motor output torque and can be used for load identification calculations.

[0036] The raw data is first filtered to remove sensor noise and occasional outliers. The preferred filtering method is sliding window mid-range filtering, with the window length selected based on signal characteristics, typically 5 to 15 sampling points. The filtered data is then normalized, mapping all data types to a unified range of 0 to 1, eliminating differences in dimensions and orders of magnitude between different physical quantities. The normalization method uses min-max normalization; for any given data value, the normalized value equals the original value minus the minimum value of that data type, divided by the difference between the maximum and minimum values.

[0037] After filtering and normalization, the system constructs a multi-dimensional feature vector for power prediction. This feature vector contains three types of information: the first type is the historical power sequence, recording the power values ​​traced back several sampling periods from the current moment. The historical window length is preferably 30 to 100 sampling points to cover the typical operation cycle of the crane. The second type is the current operating command, including the normalized values ​​of hoisting, luffing, and slewing commands. The third type is environmental parameters, including the current load mass and operating condition identifier. The construction of the multi-dimensional feature vector provides structured input data for subsequent neural network prediction.

[0038] Step S2 is dynamic power prediction. This step uses a Long Short-Term Memory (LSTM) neural network combined with an attention mechanism to predict the future power demand of each crane in the time domain. The LSTM network is a special type of recurrent neural network that solves the gradient vanishing problem in traditional recurrent neural networks when processing long sequences by introducing forget gates, input gates, and output gates. It is particularly suitable for processing power data with temporal characteristics.

[0039] Step S21 involves sequentially inputting the multidimensional feature vectors constructed in step S1 into each LSTM unit of the LSTM network. The feature vectors are input one by one in chronological order. Each LSTM unit receives the input of the current time step and the hidden state passed from the previous time step, and outputs the hidden state of the current time step after internal gate control operations. The hidden state is a compressed representation of historical information by the LSTM network, containing accumulated information from the beginning of the sequence to the current time step. After processing all time steps, a hidden state sequence is obtained, containing the hidden state vectors of each time step. The number of layers in the LSTM network and the dimensionality of the hidden state need to be determined based on the data scale and prediction accuracy requirements; the preferred hidden state dimension is 64 to 256.

[0040] Step S22 calculates the attention weights for each time step based on the hidden state sequence. The attention mechanism is a method that allows the model to automatically learn the importance of different parts of the input sequence. By assigning different weights to different time steps, the model pays more attention to historical information that is more relevant to the current prediction task during prediction. The formula for calculating the attention weights is: ; in, For the first Attention weights for each time step; It is an exponential function; For the first The attention score at each time step is calculated from the hidden state through a linear transformation and activation function; This is the summation of the attention score exponents across all time steps. The formula uses a softmax function to ensure that the sum of the attention weights across all time steps is 1, and that each weight value is between 0 and 1. The attention score is typically calculated by taking the inner product of the hidden state and the learnable parameter vector, followed by processing with a hyperbolic tangent activation function.

[0041] Step S23 involves weighted summation of the attention weights and their corresponding hidden states to obtain the context vector. The formula for calculating the context vector is: ; in, For context vectors; This represents the summation over all time steps; For the first Attention weights for each time step; For the first The hidden state vector is generated at each time step. The context vector is a weighted representation of the entire input sequence, with time steps having higher weights contributing more to the context vector. Through an attention mechanism, the model can dynamically and selectively focus on key moments in the historical sequence, such as the crane's start-up, braking, or load changes, thereby improving the specificity of the predictions.

[0042] Step S24 inputs the context vector into the fully connected output layer to obtain the power prediction sequence. The fully connected output layer consists of a linear transformation and a bias term, mapping the context vector to multi-step power prediction values. The formula for calculating the power prediction sequence is: ; in, This is a power prediction sequence, in kW; This is the output layer weight matrix; For context vectors; This is the output layer bias vector. The prediction step size is the length of the output sequence, which needs to cover the time range required for subsequent power allocation and interferometry prediction. The preferred prediction step size is 20 to 50 sampling periods, corresponding to a prediction time domain of 2 to 5 seconds.

[0043] Step S3 is online load identification. During crane operation, the actual load mass may differ from the rated value input by the operator, for example, due to errors in cargo weight estimation or changes in the spreader's own weight. By identifying the actual load mass online, the power prediction value can be corrected, improving prediction accuracy. This step uses the Kalman filter algorithm for load identification. The Kalman filter is a recursive optimal estimation algorithm that obtains the optimal state estimate in the presence of process noise and observation noise by fusing system model predictions and sensor observations.

[0044] Step S31: Establish a discrete state-space model and define the state vector. The state vector is defined as follows: ; in, for Time-state vector; for The lifting position at any given moment, in meters (m). for The lifting speed at any given time, in m / s; for The estimated load mass at any given time is expressed in kg. The state vector contains three components: position and velocity are motion state variables, and load mass is the parameter to be identified. Incorporating load mass into the state vector for joint estimation is crucial for achieving online load identification.

[0045] Step S32 predicts the current state based on the state estimate from the previous moment and the control input. The state prediction is based on the system dynamics model, considering the kinematic and dynamic characteristics of the hoisting mechanism. The state prediction formula is: ; in, for Predicted state value at any given time; This is the state transition matrix, which describes the dynamic relationships between state variables; for State estimate at time; The control input matrix describes the effect of the control input on the state. for The driving torque at any given time, expressed in N·m. State transition matrix. Based on the sampling period and the system dynamics equations, for the simplified model with the uniform velocity assumption, the next moment value of the position is equal to the current position plus the velocity multiplied by the sampling period. The velocity and load mass remain unchanged when there is no external input.

[0046] Step S33 updates the state estimate using encoder position observations. State update is the core step of Kalman filtering; by comparing the predicted position with the actual observed position, the Kalman gain is used to correct the state prediction. The state update formula is: ; in, for State estimate at time; This is the predicted state value; The Kalman gain matrix determines the weighting of predicted and observed values ​​in the final estimate. for The time and position observations, in meters, are obtained by the encoder. The observation matrix describes the relationship between state variables and observations. The Kalman gain matrix is ​​calculated based on the prediction error covariance and the observation noise covariance. A larger gain indicates higher observation reliability, meaning the state estimation relies more on the observed values; conversely, a lower gain indicates greater reliance on model predictions. State estimates are... The third component is the identified load mass. After recursive filtering over multiple sampling periods, the estimated load quality gradually converges to near the true value.

[0047] Step S34 corrects the power prediction value output in step S2 based on the identified load mass. Since the crane's lifting power is approximately proportional to the load mass, a proportional correction is needed when the identified load differs from the reference load used to train the neural network. The power prediction value correction formula is: ;

[0048] in, The power prediction is the corrected value, in kW; These are the original power prediction values, in kW; To identify the load mass, the unit is kg; The reference load mass is expressed in kg. This reference load mass represents a typical load value used during neural network training. This correction allows the power prediction model to adapt to different load conditions, improving its generalization ability.

[0049] Step S4 involves multi-crane power coordination and allocation. When multiple cranes operate simultaneously, the sum of their power demands may exceed the ship's generator power supply capacity, necessitating coordinated power allocation to optimize the power supply to each crane while ensuring safety. This step employs model predictive control (MPC) for rolling time-domain optimization. MPC is a model-based optimization control strategy that solves for the optimal control sequence within a finite time domain based on the current state and future predictions in each control cycle. However, only the first step of control is executed, and this process is repeated in the next cycle, forming rolling optimization.

[0050] Step S41 calculates the power allocation priority index for each crane. The priority index reflects the urgency of each crane's power demand; cranes with heavier loads, higher lifting heights, and greater operational urgency should receive higher priority. The priority index calculation formula is: ; in, For the first Crane priority index; This is the load weighting coefficient; For the first The current load capacity of the crane, in kg; Rated maximum load mass, in kg; This is a highly weighted coefficient; For the first The current lifting height of the crane, in meters (m). The safety height threshold is expressed in meters (m). This is the urgency weighting coefficient; For the first The emergency level of the crane operation ranges from 0 to 1; , , The sum is 1. The specific values ​​of the three weighting coefficients can be adjusted according to the actual application scenario and security requirements. The preferred values ​​are... Equal to 0.4 Equal to 0.3 Equals 0.3. Operation urgency Based on the operating range and rate of change of the master control handle, the urgency level is high when the operating range is large and the change is rapid.

[0051] Step S42 constructs the objective function for rolling time-domain optimization. The objective function contains two terms: the first term penalizes power variation to ensure smooth power allocation and avoid impacting the generator and drive system. The second term penalizes the deviation between the actual allocated power and the reference power demand, using a priority index as a weighting coefficient to ensure that higher-priority cranes receive power allocations closer to their demand. The objective function expression is: ; in, The objective function value; To sum over each time step in the prediction time domain; To sum over all cranes; Weights for power change penalties; For the first The first crane Step power change, in kW, equal to minus ; Weights for bias penalty; Priority index; For the first The first crane The reference power, in kW, is determined by the corrected power prediction value output in step S3. For the first The first crane The power allocation step, in kW, is a decision variable in the optimization solution. Two penalty weights are also included. and The value of determines the focus of optimization. When the power is larger, the power distribution is smoother. When the power requirement is large, more emphasis is placed on meeting the power demand.

[0052] Step S43 sets the optimization constraints. The constraints fall into three categories. The first category is the total power constraint, ensuring that the sum of the power of all cranes does not exceed the power available from the generator. The expression for the total power constraint is: ; in, For all cranes The sum of step power, in kW; The generator's available power, in kW, is obtained in real time from the ship's power station management system. Reserved power, measured in kW, is used to cope with sudden loads and ensure the operation of other electrical equipment on board. The value of the reserved power needs to be determined based on the ship's electrical load characteristics, and is preferably 10% to 20% of the generator's rated power. The second type is single-machine power constraint, ensuring that the power allocated to each crane is within its allowable range. The expression for the single-machine power constraint is: ; in, This is the minimum power of a single unit, measured in kW. The maximum power of a single unit is expressed in kW. The minimum power is usually set to 0 or a small positive value, while the maximum power is determined by the crane's rated power. The third type is the power change rate constraint, which limits the power change amplitude between adjacent time steps. This constraint is implicitly reflected in the objective function through a penalty term, or it can be explicitly set as a hard constraint.

[0053] Step S44 employs a quadratic programming algorithm to find the minimum value of the objective function under constraints. Since the objective function is a quadratic function of the decision variables and the constraints are linear, this optimization problem is a standard quadratic programming problem, which can be solved efficiently using the interior-point method or the effective set method. The optimal power allocation sequence for each crane within the prediction time domain is obtained. Following the rolling execution strategy of model predictive control, only the optimal power value of the first time step is output as the power allocation command for the current control cycle. In the next control cycle, the optimization problem is resolved based on the new state information. Rolling optimization enables the system to respond in real-time to environmental changes and disturbances.

[0054] Step S5 involves multi-crane interference prediction and coordinated control. When the working areas of multiple cranes overlap, it is necessary to predict the spatial interference risk between the cranes and take coordinated deceleration measures before a collision occurs. This step predicts the future spatial position of each crane based on kinematic extrapolation, calculates the minimum distance between each pair of cranes, and generates deceleration commands based on the relationship between distance and safety threshold.

[0055] Step S51 calculates the angular velocity of each crane. The angular velocity is obtained by dividing the angle difference between two adjacent sampling periods by the sampling period; this is a simple numerical differentiation method. The formula for calculating the amplitude-variable angular velocity is: ; in, For the first The luffing angular velocity of the crane, in rad / s; for The amplitude of the change in time, in rad; for The amplitude of the change in time, in rad; The sampling period is in seconds. Angular velocity. Calculate using the same method, and change the amplitude angle in the formula. Replace with rotation angle That's it. To reduce the noise impact of numerical differentiation, the angle sequence can be low-pass filtered before calculating the angular velocity.

[0056] Step S52 predicts the future spatial coordinates of each crane boom tip based on the current pose and angular velocity. The prediction uses a kinematic extrapolation method, assuming that each crane maintains its current angular velocity at a constant speed during the prediction time. The crane boom tip position is described using cylindrical coordinates, then converted to Cartesian coordinates. The boom tip spatial coordinate prediction formula contains three components: ; In the formula, , , For the first Predicted coordinates of the boom end of the gantry crane, in meters; The arm length is in meters (m). This is the current amplitude angle, in rad. The variable angular velocity is expressed in rad / s. The predicted time is in seconds. This is the current rotation angle, in rad. The rotational angular velocity is expressed in rad / s. This refers to the height of the base, in meters (m). It is a cosine function; It is a sine function. Predicted time. The selection needs to cover the time required for the crane to decelerate from its current speed to a stop, preferably 1 to 3 seconds.

[0057] Step S53 calculates the predicted distance between the boom ends of any two cranes. The Euclidean distance formula is used to calculate the straight-line distance between two points in 3D space. The formula for calculating the predicted distance between the boom ends of two cranes is: ; in, For cranes With crane Predicted distance at the end of the arm, in meters; , , For cranes Predicted coordinates of the arm tip, in meters; , , For cranes The predicted coordinates at the boom end are in meters. For multiple cranes deployed on a vessel, the predicted distances for all pairwise combinations need to be calculated. For example, for four cranes, six sets of distance values ​​need to be calculated.

[0058] Step S54 calculates the deceleration coefficient based on the relationship between the predicted distance and the safety threshold. The deceleration coefficient reflects the severity of the current interference risk, ranging from 0 to 1; a smaller coefficient indicates a higher risk and requires greater deceleration. The deceleration coefficient is calculated using a piecewise linear function. When the predicted distance... Greater than or equal to the warning distance threshold At that time, deceleration coefficient A value of 1 indicates no risk of interference, and the crane can operate at the reference speed. When the predicted distance... Greater than the parking distance threshold And less than the warning distance threshold At that time, deceleration coefficient equal This indicates that the crane has entered a warning zone and needs to reduce speed proportionally. When the predicted distance... Less than or equal to the parking distance threshold At that time, deceleration coefficient A value of 0 indicates that the crane has entered a dangerous area and must stop immediately. In the formula, This is the deceleration coefficient; The distance is predicted in meters. This is the warning distance threshold, in meters (m). The stopping distance threshold is expressed in meters. The values ​​for the warning distance and stopping distance need to be determined based on the crane's size and maximum speed; the preferred warning distance is 5m to 10m, and the stopping distance is 2m to 5m. Using a piecewise linear deceleration function instead of a step function allows for smooth deceleration, avoiding impacts on the mechanical structure and the hoisted goods from sudden stops.

[0059] Step S55 generates a coordinated deceleration command based on the deceleration coefficient and the reference speed. The formula for calculating the coordinated deceleration command is: ; in, For speed commands; For reference speed; This is the deceleration coefficient. The reference speed is determined by the operation command output from the master control handle. When there is a risk of interference between a crane and multiple cranes, the minimum value among the calculated deceleration coefficients is taken as the final deceleration coefficient for that crane, ensuring that all potential interferences are avoided.

[0060] Step S6 involves distributed communication and execution. This step distributes the power allocation command generated in step S4 and the coordinated deceleration command generated in step S5 to each crane control node for execution. The system adopts a distributed architecture, with the main controller communicating with each crane control node via a soft bus. The soft bus is a communication method based on a distributed soft bus protocol, supporting device discovery, connection establishment, and data transmission, and featuring low latency and high reliability.

[0061] The main controller encapsulates power allocation commands and coordinated deceleration commands into standard data frames. Each data frame includes a frame header to identify the start of the data frame, a device identifier to specify the target crane, a timestamp for synchronization and timeout detection, power and speed command values ​​for control data, and a checksum for data integrity verification. The encapsulated data frames are then broadcast via a soft bus or sent point-to-point to the corresponding crane control nodes.

[0062] Each crane control node receives and parses the data frames, extracting power distribution commands and coordinated deceleration commands. Based on the power distribution command, the execution control unit generates hydraulic valve control signals to adjust the hydraulic pump's displacement or the hydraulic valve's opening, thereby controlling the hydraulic system's output power. Based on the coordinated deceleration command, the execution control unit generates drive control signals to adjust the variable frequency drive's output frequency, thereby controlling the motor speed. The hydraulic system and motor drive system work together to complete the crane's motion control.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic power prediction and coordinated control of marine crane groups, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect the lifting position, luffing angle and slewing angle output by the encoder from each crane at a fixed sampling period, the hook load value output by the load sensor, the operation command value output by the master handle, and the motor current value fed back by the driver. The collected raw data is filtered to remove noise, and then normalized to map various types of data to a unified range, constructing a multi-dimensional feature vector containing historical power sequences, operation commands, and load information; S2. Dynamic Power Prediction: Input the multi-dimensional feature vector constructed in S1 into the LSTM neural network, and calculate the hidden state sequence of each time step through the LSTM unit; calculate the attention weight of each time step based on the hidden state sequence, and sum the attention weight with the hidden state to obtain the context vector; input the context vector into the output layer to output the power prediction sequence of each crane for multiple future time steps; S3. Load Online Identification: Establish a state-space model including position, velocity, and load mass. Use the Kalman filter algorithm to predict and update the state using encoder position observations. Extract the load mass estimate from the updated state vector. Correct the power prediction value output by S2 based on the ratio of the identified load mass to the reference load mass. S4. Multi-machine power collaborative allocation: Calculate the power allocation priority index based on the current load mass, lifting height and operation urgency of each crane; take the minimum power fluctuation and the minimum priority weighted deviation as the optimization objectives, and take the total generator power constraint, single machine power constraint and power change rate constraint as boundary conditions, and use the rolling time domain optimization method to solve the optimal power allocation value of each crane, and output the optimal value of the current time step as the power allocation command; S5. Multi-machine interference prediction and coordinated control: Obtain the current amplitude angle, slewing angle and boom length of each crane, and calculate the angular velocity based on the angle difference between adjacent sampling periods; predict the future spatial coordinates of each crane boom end based on the current pose and angular velocity using kinematic formulas; calculate the predicted distance between the boom ends of any two cranes, calculate the deceleration coefficient based on the relationship between the predicted distance and the warning threshold and the stopping threshold, and multiply the deceleration coefficient by the reference speed to generate a coordinated deceleration command; S6. Distributed Communication and Execution: Power allocation commands and coordinated deceleration commands are encapsulated into data frames and sent to each crane control node via a soft bus; After parsing the data frames, each control node controls the hydraulic valves to adjust the hydraulic system output according to the power distribution command, and controls the driver to adjust the motor speed according to the coordinated deceleration command.

2. The method according to claim 1, characterized in that, S2 specifically includes: S21. Input the multidimensional feature vector into each LSTM unit of the LSTM network in sequence. Each LSTM unit outputs the hidden state at the corresponding time step to obtain the hidden state sequence. S22. Calculate the attention weights at each time step using the following formula: ； in, For the first Attention weights for each time step; It is an exponential function; For the first The attention score at each time step is calculated from the hidden state through a linear transformation and activation function; Sum of the attention score index values ​​for all time steps; S23. The context vector is obtained by weighted summing of the attention weights and the hidden state, calculated using the following formula: ； in, For context vectors; This represents the summation over all time steps; For the first Attention weights for each time step; For the first The hidden state vector at each time step; S24. Input the context vector into the fully connected output layer to obtain the power prediction sequence, calculated using the following formula: ； in, For power prediction sequence (kW); This is the output layer weight matrix; For context vectors; This is the output layer bias vector.

3. The method according to claim 1, characterized in that, The load quality identification in S3 specifically includes: S31. Define the state vector ,in, for Time-state vector; for Momentary lifting position (m); for Lifting speed at any time (m / s); for Estimated load mass at any time (kg); S32. According to Time-state estimation prediction The state at time step is predicted using the following formula: ； in, for Predicted state value at any given time; This is the state transition matrix; for State estimate at time; To control the input matrix; for Constant driving torque (N·m); S33. Update the state estimate using encoder position observations. The update formula is: ； in, for State estimate at time; This is the predicted state value; The Kalman gain matrix; for Time and location observation (m); The observation matrix; The third component is the identified load quality. ; S34. Correct the power prediction value based on the identified load. The correction formula is as follows: ； in, The corrected power prediction value (kW); The original power prediction value (kW); To identify the load mass (kg); Reference load mass (kg).

4. The method according to claim 1, characterized in that, The power cooperative allocation in S4 specifically includes: S41. Calculate the priority index of each crane. The calculation formula is as follows: ； in, For the first Crane priority index; This is the load weighting coefficient; For the first Current load capacity of the crane (kg); Rated maximum load mass (kg); This is a highly weighted coefficient; For the first Current lifting height of the crane (m); The safe height threshold (m); This is the urgency weighting coefficient; For the first Emergency response level for operating a multi-stage crane; , , The sum is 1; S42. Construct the optimization objective function: ； in, The objective function value; To sum over each time step in the prediction time domain; To sum over all cranes; Weights for power change penalties; For the first The first crane Step power change (kW); Weights for bias penalty; Priority index; For the first The first crane Step reference power (kW); For the first The first crane Step-by-step power distribution (kW); S43. Set constraints, the total power constraint is: ； in, For all cranes Sum of step power (kW); The generator's available power (kW); Reserved power (kW); single unit power constraint is: ； in, This represents the minimum power output of a single unit (kW). This represents the maximum power of a single unit (kW). S44. The objective function is minimized under constraints using a quadratic programming algorithm to obtain the optimal power allocation sequence. The value of the first time step is taken as the current power allocation command output.

5. The method according to claim 1, characterized in that, The interference prediction and coordinated control in S5 specifically includes: S51. Calculate the angular velocity of each crane. The formula for calculating the luffing angular velocity is: ； in, For the first Lamp-span angular velocity of the crane (rad / s); for Time-varying angle (rad); for Time-varying angle (rad); Sampling period (s); angular velocity Calculate using the same method; S52. Predict the spatial coordinates of the arm tip. The prediction formula is as follows: ； in, , , For the first Predicted coordinates (m) of the boom end of the gantry crane; The arm length is (m). The current amplitude angle (rad); The amplitude angular velocity is (rad / s). For the predicted time (s); The current rotation angle (rad); Angular velocity (rad / s); The height of the base is (m). It is a cosine function; It is a sine function; S53. Calculate the predicted distance between the boom ends of the two cranes. The calculation formula is as follows: ； in, For cranes With crane Predicted distance at the end of the arm (m); , , For cranes Predicted coordinates of the arm tip (m); , , For cranes Predicted coordinates of the arm tip (m); S54. Calculate the deceleration coefficient based on the predicted distance: when hour, ;when hour, ;when hour, ;in, This is the deceleration coefficient; The predicted distance (m); The warning distance threshold (m); The parking distance threshold (m); S55. Generate a coordinated deceleration command; the calculation formula is as follows: ； in, For speed commands; For reference speed; This is the deceleration coefficient.

6. A dynamic power prediction and coordinated control system for marine crane fleets, used to implement the method described in any one of claims 1-5, characterized in that, include: The main controller, deployed in the ship's central control room, includes a power prediction module, a load identification module, a cooperative allocation module, an interference prediction module, and a communication management module. The power prediction module performs dynamic power prediction in step S2. The load identification module, data-connected to the power prediction module, performs load identification in step S3 and outputs correction coefficients to the power prediction module. The cooperative allocation module, data-connected to the power prediction module, performs cooperative power allocation in step S4. The interference prediction module performs interference prediction and coordinated control in step S5. The communication management module, data-connected to both the cooperative allocation module and the interference prediction module, performs data frame encapsulation and command issuance in step S6. Multiple crane control nodes are provided, each corresponding to one crane. Each crane control node is bidirectionally connected to the communication management module of the main controller via a soft bus. Each crane control node is equipped with a data acquisition unit and an execution control unit. The data acquisition unit is used to perform data acquisition and preprocessing in S1. The execution control unit is connected to the data acquisition unit and is used to perform instruction parsing and motion control in S6. Encoders: Each crane is equipped with multiple encoders, which are installed on the hoisting mechanism, luffing mechanism and slewing mechanism respectively. The signal output terminal of each encoder is connected to the position signal input terminal of the data acquisition unit of the corresponding crane control node. Each crane is equipped with one load sensor, which is installed at the hook assembly. The signal output terminal of the load sensor is connected to the load signal input terminal of the data acquisition unit of the corresponding crane control node. Each crane is equipped with a master control handle, which is installed in the operator's cab. The signal output terminal of the master control handle is connected to the command signal input terminal of the data acquisition unit of the corresponding crane control node. Hydraulic valves are provided in each crane, and the control signal input terminal of the hydraulic valve is connected to the hydraulic control output terminal of the execution control unit of the corresponding crane control node. Each crane is equipped with a driver. The control signal input terminal of the driver is connected to the drive control output terminal of the execution control unit of the corresponding crane control node, and the current feedback output terminal of the driver is connected to the current signal input terminal of the data acquisition unit of the corresponding crane control node.

7. The system according to claim 6, characterized in that, The power prediction module includes an LSTM network unit and an attention calculation unit. The input of the LSTM network unit is connected to the data acquisition unit through a communication management module to receive multi-dimensional feature vectors and output hidden state sequences. The input of the attention calculation unit is connected to the output of the LSTM network unit to calculate attention weights and output context vectors. The output of the attention calculation unit is connected to the output layer, and the output layer outputs the power prediction sequence.

8. The system according to claim 6, characterized in that, The collaborative allocation module includes a priority calculation unit, an objective function construction unit, and an optimization solution unit. The input of the priority calculation unit is connected to the data acquisition unit via a communication management module to receive the load and position information of each crane and output a priority index. The input of the objective function construction unit is connected to the output of the priority calculation unit and the output of the power prediction module. The input of the optimization solution unit is connected to the output of the objective function construction unit, and the optimization solution unit is also connected to the ship power plant management system to obtain the available power of the generator as a constraint.

9. The system according to claim 6, characterized in that, The interference prediction module includes a velocity calculation unit, a position prediction unit, and a deceleration command generation unit. The input end of the velocity calculation unit is connected to the data acquisition unit via a communication management module to receive encoder data from each crane and output angular velocity. The input end of the position prediction unit is connected to the output end of the velocity calculation unit to output predicted coordinates of each crane boom end. The input end of the deceleration command generation unit is connected to the output end of the position prediction unit to calculate the predicted distance and deceleration coefficient and output coordinated deceleration commands.

10. The system according to claim 6, characterized in that, The main controller and each crane control node are industrial control computers based on the OpenHarmony operating system; the soft bus adopts a distributed soft bus protocol, with the main controller as the master node and each crane control node as the slave node, and the master node and slave node exchange data in a publish-subscribe mode. The communication management module is also configured with a Modbus protocol interface for communication connection with the ship power plant management system.

Images