Intelligent switching and control method for ship operation mode

Abstract

This invention relates to the field of ship control technology, and more particularly to an intelligent switching and control method for ship operation modes. The method involves collecting data on ship attitude, propulsion system, operating devices, and the marine environment, and establishing a multi-dimensional state-space model that integrates the semantics of the operational task and the physical constraints of the ship. Projecting this model yields a state vector to identify the current operation mode. Based on a preset operational process and the current mode, combined with the ship's dynamic response characteristics and the coupling relationship of the operating devices, a comprehensive optimization objective function incorporating time cost and energy consumption is established to solve for the target mode and a hierarchical switching strategy. Based on the switching strategy, two-layer control commands are generated for the propulsion system and operating devices. A ship response prediction model is established to perform rolling predictions of the switching process and generate trajectory correction commands, driving the system to collaboratively complete the mode switching. This invention achieves intelligent switching and precise control of ship operation modes, improving the safety and efficiency of the switching process.

Description

Technical Field

[0001] This invention relates to the field of ship control technology, and in particular to a method for intelligent switching and control of ship operation modes. Background Technology

[0002] In the field of ship operations, mode switching and control typically rely on the operator's experience and judgment, or on decisions based on a pre-set, simple rule base. The conventional approach is to collect data from shipboard sensors and then use threshold comparisons or state machine logic to identify the current operating mode, such as switching from navigation mode to dynamic positioning mode, or from deployment / retrieval operations to towing mode. Switching strategies often employ fixed timing or manual intervention, with control commands sent independently to the propulsion system and operating equipment, lacking a holistic consideration of the coupling relationship between the two. Furthermore, the ship's trajectory during the switching process often relies solely on the current state, lacking the ability to predict short-term future dynamics.

[0003] Existing conventional practices have significant shortcomings. On the one hand, pattern recognition and switching decisions rely excessively on static rules and human experience, making it difficult to cope with complex and ever-changing sea conditions, sudden environmental disturbances, or non-standard operational tasks. This approach is not only inefficient but also prone to energy waste, such as prematurely switching to high-power propulsion modes when unnecessary or frequently adjusting the action sequence of operational devices. On the other hand, the lack of comprehensive assessment of the ship's response and the coupling effect of operational devices during the switching process makes it difficult to balance time costs and energy consumption. More importantly, existing methods fail to predict and actively correct the ship's trajectory during the switching phase. When encountering sudden changes in wind, waves, and currents, the ship may deviate from the boundary of the safe operating area, leading to risks such as collisions, capsizing, or damage to operational devices. These shortcomings restrict the efficient, safe, and intelligent switching of operational modes by ships in multi-tasking, dynamic marine environments. Summary of the Invention

[0004] The present invention provides a method for intelligent switching and control of ship operation modes, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for intelligent switching and control of ship operation modes, comprising: Collect shipboard sensor data from the target vessel, including hull attitude data, propulsion system status data, operational equipment data, and marine environment data; A multi-dimensional state space model integrating the semantics of the operation task and the physical constraints of the ship is established. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector. The current operation mode identifier is identified by combining the predefined pattern cluster. Based on the preset work process and the current work mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the work device, a comprehensive optimization objective function including time cost and energy consumption is established, and the target work mode and hierarchical switching strategy are obtained by solving it. Based on the hierarchical switching strategy, a two-level control command is generated, which includes a power allocation command for the propulsion system and an action sequence command for the working device. A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted by rolling using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and operating device to complete the mode switching in a coordinated manner.

[0006] A multi-dimensional state-space model integrating operational task semantics and ship hull physical constraints is established. The shipborne sensor data is projected and transformed within this model to obtain a state vector. This vector is then combined with a predefined pattern cluster to identify the current operational mode, including: Extract historical sensor data and corresponding operation mode labels from multiple operation processes from the historical operation database; Feature extraction is performed on the historical sensor data to obtain a set of task semantic features and a set of physical constraint features, and statistical distribution features are extracted to determine the dimensional structure of the state space and the value range of each dimension, and a multidimensional state space model is established. In the multidimensional state space model, the historical sensor data is grouped and projected according to the operation mode label, and the projection results of each group are clustered to form a predefined mode cluster. The predefined mode cluster includes the center coordinates and distribution boundary of each mode cluster. A projection transformation mapping function is constructed from the original sensor data space to the multidimensional state space model. The projection transformation mapping function achieves dimensional alignment by maintaining the correlation between features. The shipborne sensor data is input into the projection transformation mapping function to obtain the corresponding state vector in the multidimensional state space model. Calculate the spatial distance between the state vector and the center coordinates of each pattern cluster in the predefined pattern cluster, and select the pattern type corresponding to the pattern cluster with the smallest spatial distance as the current operation mode identifier.

[0007] Feature extraction is performed on the historical sensor data to obtain a task semantic feature set and a physical constraint feature set, and statistical distribution features are extracted. The dimensional structure of the state space and the value range of each dimension are determined, and a multidimensional state space model is established, including: Based on the historical sensor data, historical operation data of the working device and historical propulsion system status data are extracted. The historical operation data of the working device is segmented in time sequence to extract the device action mode. The historical propulsion system status data is analyzed by power spectrum to extract the power output frequency domain features. The device action mode and power output frequency domain features are combined to form a task semantic feature set. Historical hull attitude data and historical marine environment data are extracted from the historical sensor data. The response correlation function between hull attitude and environmental disturbance is calculated to obtain the hull attitude offset amplitude and recovery time constant corresponding to different environmental disturbance intensities. Based on the hull stability criterion, the boundary values ​​of the attitude offset amplitude and recovery time constant are determined, and the boundary values ​​constitute a set of physical constraint features. Correlation analysis is performed on the task semantic feature set to identify the dominant features and the number of the dominant features is determined as the number of task dimensions. Extreme value statistics and distribution fitting are performed on the physical constraint feature set to determine the number of physical dimensions and boundary range. Orthogonal coordinate axes are set and discrete grids are divided according to the numerical range of the dominant feature and the boundary range. The structure and range of the discrete grids are used as the dimensional structure and value range of each dimension of the multidimensional state space model.

[0008] Based on the preset work process and the current work mode identifier, and combining the ship's dynamic response characteristics and the coupling relationship of the work equipment, a comprehensive optimization objective function including time cost and energy consumption is established. Solving this function yields the target work mode and hierarchical switching strategy, including: Extract the job stage transfer rules from the preset job flow, and find the set of reachable job modes based on the current job mode identifier; Collect fuel consumption rate data and response lag data of the ship's power system under different loads, collect drive power data and switching preparation time data of the working device under different working conditions, and calculate the energy consumption rate per unit time and the time window required for mode switching. For each work mode in the set of reachable work modes, obtain the corresponding action instruction sequence, query the drive power data according to the action instruction sequence, calculate the total energy consumption in combination with the energy consumption rate per unit time, and calculate the total work time according to the execution duration of the action instruction sequence and the time window required for the mode switching. Multiply the total operation time by the time cost coefficient, multiply the total energy consumption by the energy cost coefficient, and add them together to obtain the comprehensive optimization objective function value; The operation mode corresponding to the minimum comprehensive optimization objective function value is selected as the target operation mode. The control parameters that need to be adjusted from the current operation mode to the target operation mode are identified, and a hierarchical switching strategy is generated according to the degree of influence on hull stability.

[0009] Based on the hierarchical switching strategy, a two-layer control command is generated. This two-layer control command includes a power allocation command for the propulsion system and an action sequence command for the operating device, including: The hierarchical switching strategy is analyzed to obtain the control hierarchy division result. The control hierarchy involved in the propulsion system in the control hierarchy division result is merged into the propulsion system control layer, and the control hierarchy involved in the working device is merged into the working device control layer. Extract the power demand parameters corresponding to the control layer of the propulsion system, obtain the rated power and current available power of each propulsion unit of the propulsion system, calculate the power gap between the power demand parameters and the current available power of each propulsion unit, generate the power adjustment amount of each propulsion unit based on the power gap and the load adjustment rate of each propulsion unit, and encapsulate the power adjustment amount into a power allocation instruction. Extract the task execution parameters corresponding to the control layer of the working device, obtain the motion capability range and current state position of each actuator of the working device, calculate the target state position to be reached by each actuator based on the task execution parameters, determine the motion trajectory and motion sequence of each actuator based on the current state position and the target state position, encapsulate the motion trajectory and motion sequence into action sequence instructions, and combine them with the power allocation instructions to form a two-layer control instruction.

[0010] A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted using this model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control commands are corrected to drive the propulsion system and operating device to collaboratively complete the mode switching, including: Collect attitude change data and propulsion system output data of the hull during the historical operation mode switching process, extract the dynamic response relationship between the two and establish a hull response prediction model. The ship's attitude data and propulsion system control parameters in the dual-layer control command are obtained at the current moment, and input into the ship's response prediction model to predict the ship's motion trajectory in the future time period. The ship's motion trajectory includes the ship's position sequence and attitude angle sequence. Read the boundary coordinates of the safe operating domain, compare the spatial position sequence of the ship with the boundary coordinates, identify the out-of-bounds position points that exceed the boundary, calculate the spatial deviation distance between each out-of-bounds position point and the boundary, and based on the spatial deviation distance and dynamic response relationship, solve in reverse the amount of propulsion system control parameter adjustment required to keep the ship's motion within the safe operating domain as the trajectory correction amount. The trajectory correction is superimposed on the power allocation command for the propulsion system in the dual-layer control command, while the action sequence command for the working device remains unchanged. The corrected power allocation command is sent to the propulsion system, and the action sequence command is sent to the working device.

[0011] Obtain the current hull attitude data and propulsion system control parameters from the two-layer control commands, input them into the hull response prediction model, and predict the hull trajectory over a future time period, including: Set a rolling prediction time window and a window update cycle. During mode switching, the prediction task is triggered cyclically according to the window update cycle. When each prediction task is triggered, the ship's position coordinates, the ship's three-axis attitude angles and the ship's six-degree-of-freedom velocity components are collected and the collected data are fused into the ship's attitude state. The power allocation command for the propulsion system is parsed from the two-layer control command, and the allocated power value and thrust vector direction of each propulsion unit are extracted. Based on the thrust characteristic curve of each propulsion unit, the allocated power value is converted into a thrust output value, and the thrust output value and thrust vector direction are combined into propulsion system control parameters. The hull attitude state is used as the initial state input of the hull response prediction model, and the propulsion system control parameters are used as the control input to drive the hull response prediction model to perform dynamic calculations. Within the rolling prediction time window, iterative calculations are performed according to a preset simulation step size. In each iteration, the resultant force and resultant torque on the hull are calculated based on the propulsion system control parameters, and the hull position and attitude are updated. The hull position and attitude output by each iteration step are arranged in a time sequence to generate the hull motion trajectory.

[0012] A second aspect of the present invention provides an intelligent switching and control system for ship operation modes, comprising: The data acquisition unit is used to collect data from the ship's onboard sensors, including hull attitude data, propulsion system status data, operating device operation data, and marine environment data. The state modeling unit is used to establish a multi-dimensional state space model that integrates the semantics of the operation task and the physical constraints of the ship hull. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector, and the current operation mode identifier is identified by combining a predefined pattern cluster. The optimization decision unit is used to establish a comprehensive optimization objective function that includes time cost and energy consumption based on the preset operation process and the current operation mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the operation device, and solve for the target operation mode and hierarchical switching strategy. The instruction generation unit is used to generate two-layer control instructions based on the hierarchical switching strategy. The two-layer control instructions include power allocation instructions for the propulsion system and action sequence instructions for the working device. The prediction and correction unit is used to establish a hull response prediction model, and to perform rolling prediction of the hull motion trajectory during the switching process using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operation domain, the trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and the operation device to complete the mode switching in a coordinated manner.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] This method enables intelligent identification and automatic switching of vessel operation modes, significantly improving adaptability to complex sea conditions and diverse operational tasks. By integrating multi-source data on hull attitude, propulsion system, operating devices, and marine environment, it can accurately capture the current operational status and, combined with preset operational procedures and hull dynamic characteristics, intelligently generate the optimal target mode and hierarchical switching strategy.

[0016] This method optimizes energy consumption and time costs during mode switching. The comprehensive optimization objective function takes into account the ship's dynamic response characteristics and the coupling relationship between the operating devices. The resulting hierarchical switching strategy minimizes unnecessary energy losses in the propulsion system and operating devices while meeting operational efficiency requirements. Compared with traditional fixed switching logic, this method can dynamically adjust power allocation and action sequence according to real-time operating conditions, enabling all systems to operate collaboratively within the optimal energy efficiency range, thereby extending equipment lifespan and reducing operating costs.

[0017] This method significantly improves the safety and accuracy of mode switching through the rolling prediction and trajectory correction mechanism of the hull response prediction model. This closed-loop control strategy effectively suppresses the impact load and attitude deviation during the switching process, preventing hull instability or collision of working equipment. It is especially suitable for scenarios with extremely high safety and accuracy requirements, such as deep-sea mining and offshore wind power installation, and provides reliable protection for unmanned and intelligent ship operations. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent switching and control method for ship operation modes according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the method for identifying the current work mode identifier according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the intelligent switching and control method for ship operation modes according to an embodiment of the present invention. The present invention provides an intelligent switching and control method for ship operation modes, including: Collect shipboard sensor data from the target vessel, including hull attitude data, propulsion system status data, operational equipment data, and marine environment data; A multi-dimensional state space model integrating the semantics of the operation task and the physical constraints of the ship is established. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector. The current operation mode identifier is identified by combining the predefined pattern cluster. Based on the preset work process and the current work mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the work device, a comprehensive optimization objective function including time cost and energy consumption is established, and the target work mode and hierarchical switching strategy are obtained by solving it. Based on the hierarchical switching strategy, a two-level control command is generated, which includes a power allocation command for the propulsion system and an action sequence command for the working device. A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted by rolling using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and operating device to complete the mode switching in a coordinated manner.

[0022] Figure 2 This is a flowchart illustrating the method for identifying the current operation mode identifier according to an embodiment of the present invention. A multi-dimensional state space model integrating operation task semantics and ship hull physical constraints is established. The shipborne sensor data is projected and transformed within the multi-dimensional state space model to obtain a state vector. The current operation mode identifier is then identified by combining this vector with a predefined pattern cluster. The process includes: Extract historical sensor data and corresponding operation mode labels from multiple operation processes from the historical operation database; Feature extraction is performed on the historical sensor data to obtain a set of task semantic features and a set of physical constraint features, and statistical distribution features are extracted to determine the dimensional structure of the state space and the value range of each dimension, and a multidimensional state space model is established. In the multidimensional state space model, the historical sensor data is grouped and projected according to the operation mode label, and the projection results of each group are clustered to form a predefined mode cluster. The predefined mode cluster includes the center coordinates and distribution boundary of each mode cluster. A projection transformation mapping function is constructed from the original sensor data space to the multidimensional state space model. The projection transformation mapping function achieves dimensional alignment by maintaining the correlation between features. The shipborne sensor data is input into the projection transformation mapping function to obtain the corresponding state vector in the multidimensional state space model. Calculate the spatial distance between the state vector and the center coordinates of each pattern cluster in the predefined pattern cluster, and select the pattern type corresponding to the pattern cluster with the smallest spatial distance as the current operation mode identifier.

[0023] In practical applications, data from multiple operations covering different sea states and task types are extracted from the ship's historical operation database. This historical data includes roll, pitch, and heading angles and their angular velocities recorded by hull attitude sensors; main engine speed, propeller thrust, and rudder angle feedback values ​​recorded by propulsion system status sensors; crane load, winch tension, and hydraulic system pressure recorded by operational device sensors; and wave height, wave period, wind speed, and current velocity recorded by marine environment sensors. Each set of historical sensor data is associated with manually labeled or automatically recorded operation mode tags, such as "anchoring operation mode," "dynamic positioning mode," "towing operation mode," and "crane operation mode." Data extraction must ensure timestamp alignment and a consistent sampling frequency of 1Hz to 10Hz to guarantee temporal consistency in subsequent feature extraction.

[0024] When extracting features from historical sensor data, the raw sensor signals are divided into two feature sets: a task semantic feature set reflecting the essential attributes of the task, including the combination of working states of the working devices, the coordination mode between the propulsion system and the working devices, and the spatial relationship between the ship and the target. For example, in lifting operations, if the crane load exceeds a set threshold and the winch is locked, while the propulsion system maintains low-speed micro-movement, these features are combined to form the semantic features of this mode. The physical constraint feature set reflects the ship's dynamic constraints and safety boundaries, including the upper limit of the rate of change of the ship's attitude angle, the physical limit of the propulsion system's power output, and the safety margin of the working device load. Statistical distribution characteristics are obtained by calculating the mean, variance, extreme values, and quantiles of each feature in the historical data. The dimensional structure of the state space is determined based on the variance contribution rate and correlation analysis of the features. If the variance contribution rate of a feature is less than 2% or the absolute value of its correlation coefficient with other features is greater than 0.95, it is removed or merged, ultimately retaining 8 to 15 independent dimensions. The value range for each dimension is determined based on three times the standard deviation of historical data to ensure coverage of all possible states under normal operating conditions.

[0025] The established multidimensional state-space model employs normalization processing, mapping the values ​​of each dimension to a standard range of 0 to 1. In this space, each dimension corresponds to a specific physical quantity or semantic attribute; for example, the first dimension represents the normalized value of the ship's roll angle, the second dimension represents the normalized value of the main engine power output, and the third dimension represents the normalized value of the crane load. Historical sensor data is grouped according to operation mode labels, with each group corresponding to one operation mode. A projection transformation is performed on each group of data, mapping the original high-dimensional sensor data to the multidimensional state-space model, forming the point set distribution of that group of data in the state space. K-means clustering or Gaussian mixture model is used to perform cluster analysis on the projection results of each group, identifying the typical distribution area of ​​each operation mode in the state space. During clustering, the number of clusters is set to be consistent with the number of operation mode labels. Iterative optimization ensures that data points of the same mode are clustered in the space, while data points of different modes are separated. After clustering, the center coordinates of each mode cluster are calculated, i.e., the mean vector of coordinates of all data points within that cluster. At the same time, the distribution boundary is determined based on the distribution of data points within the cluster. Typically, a hyperellipsoid or convex hull containing 95% of the data points is used as the boundary description.

[0026] When constructing the projection transformation mapping function from the original sensor data space to a multi-dimensional state space model, principal component analysis or kernel function mapping methods are employed. The projection transformation mapping function must preserve the correlation structure between features to avoid information loss. In specific implementation, the original sensor data is standardized to eliminate the influence of dimensions. The standardized data is then mapped to various dimensions of the state space through linear transformation or a nonlinear kernel function. The linear transformation is implemented using an eigenvector matrix, which is obtained by decomposing the covariance matrix of historical data. The nonlinear kernel function can be a radial basis function or a polynomial kernel function, and the degree of nonlinearity of the mapping is controlled by adjusting the kernel parameters. During the projection transformation process, it is ensured that the task semantic features and physical constraint features are dimensionally aligned in the state space; that is, features of the same type are mapped to the same dimensional interval.

[0027] In actual operation, the currently collected shipborne sensor data is input into the projection transformation mapping function. The sensor data needs to be preprocessed, including outlier removal, filtering and noise reduction, and time alignment. The preprocessed data is normalized using the same standardization method as historical data, and the state vector is calculated through the projection transformation mapping function. The state vector is a multidimensional array, where each element corresponds to the coordinate value of each dimension in the state space model. For example, if the state space is 12-dimensional, the state vector contains 12 values, representing the normalized state of the current ship in each feature dimension.

[0028] When calculating the spatial distance between the state vector and the center coordinates of each predefined pattern cluster, Euclidean distance or Mahalanobis distance is used as the metric. Euclidean distance is simple to calculate and suitable for situations where each dimension is independent and has similar variances. Mahalanobis distance considers the covariance structure of each dimension and can more accurately reflect the similarity between the state vector and the pattern cluster. For each predefined pattern cluster, the distance value between the state vector and the center coordinates of that cluster is calculated. After traversing all pattern clusters, the pattern cluster with the smallest distance value is selected. The pattern type corresponding to this pattern cluster is the current operation mode identifier. For example, if the distance between the state vector and the center of the "Dynamic Positioning Mode" cluster is 0.15, the distance between the state vector and the center of the "Anchoring Operation Mode" cluster is 0.42, and the distance between the state vector and the center of the "Towing Operation Mode" cluster is 0.68, then the current operation mode is identified as "Dynamic Positioning Mode".

[0029] To improve the robustness of identification, a confidence assessment mechanism is introduced based on distance calculation. If the distance between the state vector and the center of the nearest pattern cluster exceeds the threshold of the cluster's distribution boundary, the current state is determined to be a transitional or abnormal state. Instead of directly outputting the pattern identifier, a secondary confirmation process is triggered. This secondary confirmation process analyzes the trajectory of the state vector over multiple consecutive sampling periods to determine whether the state vector stably converges into a certain pattern cluster. If the state vector falls within the distribution boundary of the same pattern cluster for five consecutive sampling periods, the pattern is confirmed as the current operating mode identifier. If the state vector hovers near the boundaries of multiple pattern clusters, a temporary "mode switching in progress" identifier is output, and final identification is performed after the state stabilizes.

[0030] In the selection of dimensions for multidimensional state-space models, task semantic features typically occupy 3 to 5 dimensions, reflecting the working state of the operating device, the operation mode of the propulsion system, and the relative positional relationship between the hull and the operating target. Physical constraint features occupy 5 to 10 dimensions, including the stability index of the hull attitude, the power margin of the propulsion system, and the load safety factor of the operating device. The determination of the number of dimensions requires a trade-off between model complexity and computational efficiency. Too many dimensions will increase the computational burden and the risk of overfitting, while too few dimensions will not be able to fully represent the differences in operating modes.

[0031] The number of predefined pattern clusters corresponds to the number of operational mode types supported by the vessel. For multi-functional vessels, there will be 8 to 12 different operational modes, corresponding to 8 to 12 pattern clusters. The distribution boundary of each pattern cluster is determined through statistical analysis of historical data. The boundary shape can be a hypersphere, a hyperellipsoid, or an irregular convex hull, depending on the actual distribution of the pattern in the state space. There are some overlapping regions between pattern clusters. For state vectors falling within the overlapping regions, discrimination is made by comparing their distances to the centers of each cluster and combining historical transition probabilities.

[0032] The parameters of the projection transformation mapping function are obtained through offline training, using all data from the historical operation database. The training objective is to minimize the classification error between the projected state vector and the actual operation mode label, while maximizing the spatial separation between different mode clusters. After training, the parameters of the mapping function are permanently stored and can be directly called during actual runtime without retraining. As new operation data accumulates, the model can be updated periodically, retraining the projection transformation mapping function and updating the center coordinates and distribution boundaries of predefined mode clusters to adapt to changes in ship operation characteristics or the needs of new operation modes.

[0033] Feature extraction is performed on the historical sensor data to obtain a task semantic feature set and a physical constraint feature set, and statistical distribution features are extracted. The dimensional structure of the state space and the value range of each dimension are determined, and a multidimensional state space model is established, including: Based on the historical sensor data, historical operation data of the working device and historical propulsion system status data are extracted. The historical operation data of the working device is segmented in time sequence to extract the device action mode. The historical propulsion system status data is analyzed by power spectrum to extract the power output frequency domain features. The device action mode and power output frequency domain features are combined to form a task semantic feature set. Historical hull attitude data and historical marine environment data are extracted from the historical sensor data. The response correlation function between hull attitude and environmental disturbance is calculated to obtain the hull attitude offset amplitude and recovery time constant corresponding to different environmental disturbance intensities. Based on the hull stability criterion, the boundary values ​​of the attitude offset amplitude and recovery time constant are determined, and the boundary values ​​constitute a set of physical constraint features. Correlation analysis is performed on the task semantic feature set to identify the dominant features and the number of the dominant features is determined as the number of task dimensions. Extreme value statistics and distribution fitting are performed on the physical constraint feature set to determine the number of physical dimensions and boundary range. Orthogonal coordinate axes are set and discrete grids are divided according to the numerical range of the dominant feature and the boundary range. The structure and range of the discrete grids are used as the dimensional structure and value range of each dimension of the multidimensional state space model.

[0034] Before constructing a multidimensional state-space model, it is necessary to extract key information reflecting the essential characteristics of the operational tasks and the physical limitations of the hull from the historical operational records of the target vessel. Historical sensor data is usually stored in time-series format, covering complete records of multiple operational cycles. By systematically processing this data, typical characteristic patterns under different operational modes can be revealed, and the safe operating boundaries of the hull under various environmental conditions can be determined.

[0035] Historical operational data of the working device and historical status data of the propulsion system are separated from historical sensor data. The historical operational data includes time series of parameters such as crane boom angle, winch speed, hydraulic system pressure, and tool position. When segmenting these time series, a sliding window method is used to identify steady-state and transitional segments. Specifically, the time window length is set from 30 to 120 seconds, adjusted according to the response speed characteristics of the working device. Within each time window, the mean, variance, and rate of change of each parameter are calculated. When the statistical characteristics remain stable across multiple consecutive windows, it is determined to be a steady-state operational segment; when a significant jump occurs in the statistical characteristics, it is determined to be a transitional segment of mode switching.

[0036] By performing cluster analysis on parameter combinations during steady-state periods, the device's operational modes can be extracted. For example, in the dynamic positioning operations of marine engineering vessels, typical operational modes such as "heavy-load lifting," "empty-load recovery," and "standby hovering" can be identified. Each operational mode corresponds to a specific set of parameter characteristics. For instance, in the heavy-load lifting mode, the boom angle is typically between 15 and 45 degrees, the hydraulic system pressure is maintained at a high level, and the winch is in a low-speed, stable state. These parameter characteristics are encoded into feature vectors, forming a digital representation of the device's operational modes.

[0037] Power spectrum analysis was performed on historical propulsion system status data to reveal the frequency domain characteristics of power output. The propulsion system status data included parameters such as main engine speed, propeller thrust, azimuth angle, and power output. A Fast Fourier Transform (FFT) was used to convert the time-domain signal to the frequency domain, yielding the power spectral density function. In the frequency domain, the power output under different operating modes exhibited characteristic spectral distributions. For example, in towing operation mode, the propulsion system power output was mainly concentrated in the low-frequency band, corresponding to stable continuous thrust; while in dynamic positioning mode, the power output showed significant energy distribution in the mid-to-high frequency band, reflecting frequent thrust adjustments. Characteristic parameters such as peak frequency, main bandwidth, and energy concentration of the power spectrum were extracted as quantitative indicators of the frequency domain characteristics of power output.

[0038] By combining the device's operational modes and power output frequency domain features, a task semantic feature set is constructed. This set typically contains features with 8 to 20 dimensions, the exact number depending on the vessel type and operational complexity. For complex engineering vessels equipped with multiple sets of operational devices, the feature dimensions can reach over 30 dimensions. The task semantic feature set can describe the vessel's operational status from a functional perspective, providing a semantic foundation for subsequent pattern recognition.

[0039] Historical ship attitude data and historical marine environment data are extracted from historical sensor data. The historical ship attitude data includes six-degree-of-freedom motion parameters such as roll angle, pitch angle, bow angle, heave displacement, sway displacement, and pitch displacement. The historical marine environment data includes environmental disturbance parameters such as wind speed, wind direction, wave height, wave period, current speed, and current direction. These two types of data are aligned according to timestamps to establish the correspondence between environmental disturbances and ship response.

[0040] The correlation function between hull attitude and environmental disturbance response is calculated, and cross-correlation analysis is used to quantify the causal relationship between the two. For specific types of environmental disturbances, such as lateral wind loads, the cross-correlation coefficient between the disturbance and the roll angle is calculated as a function of time delay. The peak value of the cross-correlation coefficient corresponds to the moment of maximum hull response to the disturbance, and the magnitude of the peak value reflects the intensity of the response. By grouping and statistically analyzing environmental disturbances of different intensities, the mapping relationship between disturbance intensity and hull attitude deviation amplitude is obtained. For example, when the lateral wind speed increases from 5 m / s to 15 m / s, the roll angle amplitude increases from 2 degrees to 8 degrees.

[0041] In addition to the offset amplitude, the recovery time constant also needs to be extracted to characterize the speed at which the hull recovers to its equilibrium state after the disturbance disappears. The time response curve of the hull attitude is fitted with an exponential decay model, and the decay time constant is extracted. The recovery time constant reflects the damping characteristics and recovery capability of the hull and is an important indicator for evaluating hull stability. The recovery time constant varies significantly for different environmental disturbance intensities and hull loading conditions. Through statistical analysis, a multiple regression model is established between environmental disturbance intensity, hull loading condition, and recovery time constant.

[0042] The boundary values ​​for attitude deviation amplitude and recovery time constant are determined based on hull stability criteria. Hull stability criteria include the complete stability rules and damaged stability rules established by the International Maritime Organization (IMO), as well as technical specifications issued by classification societies. These criteria specify the permissible range of hull attitude under different sea states. For example, for marine engineering vessels, the roll angle should generally not exceed 15 degrees, and the pitch angle should not exceed 10 degrees. These boundary values ​​are further tightened in conjunction with operational safety requirements. In precision operation mode, the roll angle needs to be limited to within 5 degrees to ensure the positioning accuracy of the operating equipment. These boundary values ​​are compared with statistically obtained actual response data to determine the boundaries of the safe operating domain. The boundary value of the recovery time constant is determined based on operational continuity requirements; excessively long recovery times can lead to operational interruptions or decreased efficiency. The attitude deviation amplitude boundaries and recovery time constant boundaries constitute a set of physical constraint features, which typically contains 6 to 12-dimensional features, corresponding to the amplitude and time constraints of six-degree-of-freedom motion.

[0043] Correlation analysis is performed on the task semantic feature set to identify the dominant features that contribute most to distinguishing operation modes. Principal component analysis or mutual information methods are used to evaluate the correlation and redundancy among the features. Highly correlated features carry redundant information; only the most representative features are retained. The contribution rate of each feature to the total variance is calculated, and features are sorted from highest to lowest contribution rate. Features with a cumulative contribution rate of 85% to 95% are selected as dominant features. The number of dominant features corresponds to the number of task dimensions, typically between 5 and 15. For example, for marine engineering vessels, dominant features include boom angle, hydraulic pressure, main engine power, thrust distribution variance, and tool speed. These features effectively distinguish different operation modes while avoiding computational complexity issues caused by excessive dimensionality.

[0044] Extreme value statistics and distribution fitting are performed on the physical constraint feature set to determine the number of physical dimensions and the boundary range. For each physical constraint feature, its maximum, minimum, mean, and standard deviation in historical data are statistically analyzed. Kernel density estimation or parametric distribution fitting methods are used to obtain the probability distribution function of each feature. Commonly used distribution models include the normal distribution, Weibull distribution, and extreme value distribution. The most suitable distribution model is selected through goodness-of-fit testing. Based on the fitted distribution function, the effective range of feature values ​​is determined. Typically, an interval covering 99% of the probability is selected as the effective range to exclude the influence of extreme outliers. The number of physical dimensions is equal to the number of independent features in the physical constraint feature set, generally 6 to 12 dimensions. The boundary range is determined jointly by the stability criterion and the statistical distribution; the more stringent constraint is taken as the final boundary.

[0045] Orthogonal coordinate axes are established to combine the task dimension and physical dimension into a multidimensional state space. The task dimension reflects the functional characteristics of the operation, while the physical dimension reflects the safety constraint characteristics. These two types of dimensions are semantically independent, thus allowing the construction of an orthogonal coordinate system. The number of coordinate axes equals the sum of the number of task dimensions and the number of physical dimensions, typically ranging from 12 to 25 dimensions. Discrete grids are created based on the numerical range and boundary extent of the dominant features. For each dimension, the grid density is determined based on the feature distribution characteristics and practical application requirements. For dimensions with drastic changes or significant impacts on pattern recognition, a denser grid is used, with 50 to 100 grids; for dimensions with gradual changes, a sparser grid is used, with 10 to 30 grids. Grid partitioning can employ equal spacing or adaptive partitioning based on quantiles. Adaptive partitioning provides higher resolution in data-dense regions and reduces computational burden in data-sparse regions.

[0046] The structure and extent of the discrete grid serve as the dimensional structure and value range of each dimension in the multidimensional state-space model. The grid structure defines the topological relationships of the state space, with each grid cell representing a discrete state. The total number of states in the state space equals the product of the number of grid cells in each dimension. For a 15-dimensional space with an average of 30 grid cells per dimension, the total number of states can reach billions. To reduce computational complexity, a sparse representation method can be used, retaining only states that have actually appeared in historical data and their neighborhoods, ignoring unvisited state regions. The value range of each dimension is determined by the grid boundaries. Sensor data exceeding the range will be mapped to the boundary grid during projection transformation or marked as an abnormal state to trigger an alarm. After the multidimensional state-space model is established, it can be used for projection transformation of real-time sensor data and operation mode recognition, providing a state representation basis for subsequent mode switching decisions.

[0047] Based on the preset work process and the current work mode identifier, and combining the ship's dynamic response characteristics and the coupling relationship of the work equipment, a comprehensive optimization objective function including time cost and energy consumption is established. Solving this function yields the target work mode and hierarchical switching strategy, including: Extract the job stage transfer rules from the preset job flow, and find the set of reachable job modes based on the current job mode identifier; Collect fuel consumption rate data and response lag data of the ship's power system under different loads, collect drive power data and switching preparation time data of the working device under different working conditions, and calculate the energy consumption rate per unit time and the time window required for mode switching. For each work mode in the set of reachable work modes, obtain the corresponding action instruction sequence, query the drive power data according to the action instruction sequence, calculate the total energy consumption in combination with the energy consumption rate per unit time, and calculate the total work time according to the execution duration of the action instruction sequence and the time window required for the mode switching. Multiply the total operation time by the time cost coefficient, multiply the total energy consumption by the energy cost coefficient, and add them together to obtain the comprehensive optimization objective function value; The operation mode corresponding to the minimum comprehensive optimization objective function value is selected as the target operation mode. The control parameters that need to be adjusted from the current operation mode to the target operation mode are identified, and a hierarchical switching strategy is generated according to the degree of influence on hull stability.

[0048] After determining the current operating mode identifier, it is necessary to further determine which target operating mode the vessel should switch to and plan a reasonable switching path. This process requires comprehensive consideration of the logical constraints of the operating process, the physical characteristics of the vessel's power system, and the coupling relationship between the operating devices, seeking a balance between time efficiency and energy economy.

[0049] The operation phase transition rules are extracted from the preset operation flow, which is usually stored in the form of a directed graph. Nodes in the graph represent different operation modes, and edges represent allowed mode transition paths. Each edge is labeled with transition conditions; for example, "anchoring mode" can only transition to "slow navigation mode" or "dynamic positioning mode," but cannot directly jump to "high-speed navigation mode." Based on the current operation mode identifier, the current node is located in the directed graph. Breadth-first search or depth-first search algorithms are used to find all nodes reachable from the current node. The operation modes corresponding to these nodes constitute the set of reachable operation modes. During the search process, it is necessary to check whether the transition conditions on each edge are met. For example, some transitions require the sea state level to be below a specific threshold, or require specific operational equipment to be in a standby state. In this way, operation modes that cannot be directly or indirectly reached under the current conditions can be excluded, narrowing the scope of subsequent optimization calculations.

[0050] Data on fuel consumption rate and response lag of the ship's propulsion system under different loads were collected. The ship's propulsion system typically includes main engines, auxiliary engines, and propellers, and its fuel consumption rate exhibits a non-linear relationship with output power. By conducting actual ship tests under different load conditions or referring to performance curves provided by equipment manufacturers, a series of discrete power-fuel consumption data points were obtained. For example, the fuel consumption rate is 185 grams per kilowatt-hour at 30% load and 210 grams per kilowatt-hour at 70% load. These data points were interpolated and fitted to obtain a continuous fuel consumption rate function. Response lag data reflects the time required for the propulsion system to reach the target output value from receiving a command. This time is affected by factors such as mechanical inertia and the response speed of the hydraulic system. For example, it takes 15 to 30 seconds for the main engine to accelerate from idle speed to rated speed, while it takes 8 to 12 seconds for the propeller to increase from zero thrust to maximum thrust. This response lag data is crucial for evaluating the time cost of mode switching.

[0051] Data on drive power and switching preparation time of the operating devices under different working conditions is collected. These devices include cranes, winches, drilling equipment, and dynamic positioning systems, each with significantly different power requirements under different conditions. For example, a crane requires only 5 kW when slewing under no-load conditions, but up to 150 kW when lifting under full load. A mapping table between operating conditions and power requirements is established by consulting the device's technical manual or actual measurement data. Switching preparation time data refers to the time required for the device to transition from one working state to another. For example, it takes 120 seconds for a crane to extend from its retracted state to its working position, and 5 seconds for a winch to release from its braking state to its operational state. This time data must account for the physical limitations of mechanical movements and cannot be artificially compressed.

[0052] Calculate the energy consumption rate per unit time. Convert the fuel consumption rate data of the ship's propulsion system into an energy consumption rate, considering the calorific value of the fuel. For example, the calorific value of marine diesel is approximately 42 MJ / kg, so the energy consumption rate corresponding to a fuel consumption rate of 200 g / kW is approximately 2.33 MJ / s. Simultaneously, sum the drive power data of the operating devices to obtain the total power requirement of all devices under a specific operating condition. Assuming that in a certain operating mode the main engine output power is 800 kW, the auxiliary engine output power is 200 kW, the crane power requirement is 100 kW, and the dynamic positioning system power requirement is 150 kW, then the total power requirement is 1250 kW. Using the fuel consumption rate function, calculate the energy consumption rate per unit time under this operating condition.

[0053] The time window required for mode switching is calculated, and it consists of two parts: the power system response time and the preparation time for switching the working device. For example, switching from "anchoring mode" to "dynamic positioning mode" requires starting the dynamic positioning system, which has an initialization time of 30 seconds. Simultaneously, the host machine needs to be upgraded from idle speed to medium load, with a response time of 20 seconds. Since these two actions can be performed in parallel, the larger of the two values, 30 seconds, is used for the time window. If the anchor chain also needs to be retrieved, and this retrieval operation takes 180 seconds and must be completed before the dynamic positioning system starts, then the time window is increased to 210 seconds. The accurate time window is calculated by analyzing the dependencies and parallel possibilities between the actions.

[0054] For each operating mode in the reachable operating mode set, the corresponding action command sequence is obtained. Each operating mode corresponds to a specific set of equipment state configurations. Switching from the current mode to the target mode requires executing a series of action commands. For example, switching from "slow navigation mode" to "operating mode" includes the following action command sequence: reducing the main engine speed to 30% of the rated speed, activating the dynamic positioning system, deploying the operating device, and adjusting the thruster angle to the positioning mode. Each command is associated with an execution duration and power requirement. Based on the action command sequence, drive power data is queried to obtain the power requirement curve for each time period.

[0055] The total energy consumption is calculated by combining the energy consumption rate per unit time. The sequence of action commands is divided into several time periods, and the power demand remains relatively stable within each time period. For the first time period... The duration of each time period is The average power requirement is The corresponding fuel consumption rate is The energy consumption during that time period is The total energy consumption is obtained by summing up the energy consumption over all time periods. ,in This represents the total number of time periods. This calculation process needs to consider the transient characteristics of the power system under varying load conditions; for example, the fuel consumption rate during acceleration is usually higher than that under steady-state conditions.

[0056] The total job time is calculated based on the execution duration of the action instruction sequence and the time window required for mode switching. The execution duration of the action instruction sequence is the sum of the times of all sequential actions, while also considering the overlapping parts of parallel actions. Assuming a switching process consists of three sequential phases: Phase 1 lasts 40 seconds, Phase 2 lasts 60 seconds, and Phase 3 lasts 30 seconds, then the execution duration is 130 seconds. The time window required for mode switching reflects the shortest time required for the system to switch from initial setup to complete stability. If the execution duration is less than the time window, the total job time is the time window value; otherwise, the execution duration is used. In addition, the buffer time required for the system to stabilize after the switch must be added; for example, a dynamic positioning system requires an additional 20 seconds to reach a stable positioning state.

[0057] Multiply the total operation time by the time cost coefficient, and multiply the total energy consumption by the energy cost coefficient, then add them together to obtain the comprehensive optimization objective function value. The time cost coefficient reflects the economic value per unit of time; for example, the time cost coefficient is higher in emergency operation scenarios and lower in routine operation scenarios. The energy cost coefficient reflects the economic value per unit of energy and is usually related to fuel prices. Assume the time cost coefficient is... The energy cost coefficient is The total work time is Total energy consumption is The comprehensive optimization objective function value is By adjusting and The relative size of the energy source can be used to balance time efficiency and energy economy.

[0058] The operation mode corresponding to the minimum comprehensive optimization objective function value is selected as the target operation mode. For each operation mode in the set of reachable operation modes, its corresponding comprehensive optimization objective function value is calculated, and the operation mode with the smallest function value is found by comparison. Assuming that the set of reachable operation modes includes three modes: "Dynamic Positioning Mode", "Slow Navigation Mode" and "Anchor-Assisted Mode", with comprehensive optimization objective function values ​​of 1250, 1180 and 1320 respectively, then "Slow Navigation Mode" is selected as the target operation mode. This selection process is essentially a discrete optimization problem. When the number of reachable operation modes is large, a heuristic algorithm can be used to accelerate the solution.

[0059] Identify the control parameters that need adjustment when switching from the current operating mode to the target operating mode. By comparing the equipment status configuration between the current and target operating modes, identify the control parameters that need to be changed. For example, if the main engine speed is 600 rpm in the current mode and 450 rpm in the target mode, then the main engine speed is the control parameter that needs adjustment. Similarly, parameters such as propeller angle, working device position, and power distribution ratio also need to be adjusted. List these control parameters to create a parameter adjustment list.

[0060] A tiered switching strategy is generated based on the degree of impact on hull stability, as adjustments to different control parameters have varying degrees of influence on hull stability. For example, a sudden change in propeller thrust can cause significant roll or pitch, while adjustments to auxiliary engine power have a smaller impact on hull attitude. A sensitivity matrix is ​​established between control parameters and hull stability indices to quantify the impact of each parameter adjustment on stability indicators such as roll, pitch, and heave. Control parameters are sorted from least to most influential, prioritizing adjustments to parameters with smaller impacts and ending with those with larger impacts. For example, the switching strategy might be divided into three tiers: the first tier adjusts auxiliary engine power and the lighting system; the second tier adjusts main engine speed and the position of the operating devices; and the third tier adjusts propeller angle and power distribution ratio. Parameter adjustments within each tier can be performed in parallel, while adjustments between different tiers are performed sequentially to ensure the hull remains within a safe and stable region throughout the switching process.

[0061] Based on the hierarchical switching strategy, a two-layer control command is generated. This two-layer control command includes a power allocation command for the propulsion system and an action sequence command for the operating device, including: The hierarchical switching strategy is analyzed to obtain the control hierarchy division result. The control hierarchy involved in the propulsion system in the control hierarchy division result is merged into the propulsion system control layer, and the control hierarchy involved in the working device is merged into the working device control layer. Extract the power demand parameters corresponding to the control layer of the propulsion system, obtain the rated power and current available power of each propulsion unit of the propulsion system, calculate the power gap between the power demand parameters and the current available power of each propulsion unit, generate the power adjustment amount of each propulsion unit based on the power gap and the load adjustment rate of each propulsion unit, and encapsulate the power adjustment amount into a power allocation instruction. Extract the task execution parameters corresponding to the control layer of the working device, obtain the motion capability range and current state position of each actuator of the working device, calculate the target state position to be reached by each actuator based on the task execution parameters, determine the motion trajectory and motion sequence of each actuator based on the current state position and the target state position, encapsulate the motion trajectory and motion sequence into action sequence instructions, and combine them with the power allocation instructions to form a two-layer control instruction.

[0062] After obtaining the hierarchical switching strategy, it needs to be translated into control commands that can directly drive the ship's hardware system. Since the ship's propulsion system and operating devices differ significantly in physical characteristics, response time, and control logic, a two-tiered control architecture can achieve refined collaborative control.

[0063] The hierarchical switching strategy is analyzed to extract the control hierarchy division information. This strategy typically contains multiple control levels, each corresponding to a specific controlled object and control timing. By traversing all hierarchical nodes in the strategy, the type of hardware object associated with each node is identified. If a node involves propulsion-related equipment such as the main thruster, side thrusters, or servo motors, it is classified into the propulsion system control layer; if it involves working devices such as cranes, winches, hydraulic booms, or work platforms, it is classified into the working device control layer. This merging operation is essentially a classification and aggregation of controlled objects, enabling subsequent instruction generation to be customized for the characteristics of different hardware systems.

[0064] For the propulsion system control layer, power allocation commands need to be generated, and power demand parameters need to be extracted from this control layer. These parameters include target speed, target heading, and required thrust vector. Simultaneously, the technical parameters of each propulsion unit in the propulsion system need to be obtained, including the main engine's rated power, the side thrusters' rated power, and the current operating power of each unit. The current available power needs to consider the real-time status of the equipment; for example, if a main engine is undergoing maintenance, its available power is zero, or if a propulsion unit is operating at reduced capacity due to overheating.

[0065] When calculating the power gap, the power demand parameters are converted into a total power demand and compared with the sum of the currently available power of each propulsion unit. If the total power demand exceeds the sum of the currently available power, a power gap exists, requiring the activation of backup propulsion units or adjustment of the work plan. If the currently available power is sufficient, it needs to be rationally allocated among the propulsion units. Power allocation needs to consider the load adjustment rate of each unit, which reflects the time required for the propulsion unit to adjust from its current power state to the target power state. For example, the power adjustment rate of a diesel engine is typically slow, requiring tens of seconds to complete a significant power adjustment; while the response speed of an electric propulsion unit is faster, completing a power adjustment within seconds.

[0066] Based on the load adjustment rate and current power status of each propulsion unit, the power adjustment amount for each unit is calculated. Units with fast response times can be assigned larger power adjustment tasks; for units with slow response times, the adjustment process needs to be initiated earlier or a smaller adjustment range needs to be assigned. The calculation of the power adjustment amount also needs to consider the efficiency characteristics of each unit, prioritizing high-efficiency units to bear more load. The calculated power adjustment amounts for each propulsion unit are encapsulated in a standardized format to form a power allocation instruction. This instruction contains fields such as target power value, adjustment rate, and priority, and can be directly parsed and executed by the underlying controller of the propulsion system.

[0067] For the control layer of the working device, it is necessary to generate action sequence instructions and extract task execution parameters from this control layer. These parameters describe the specific actions that the working device needs to perform. For example, the crane needs to move the hook from position A to position B, the winch needs to wind up or down the cable to a specific length, and the hydraulic arm needs to be adjusted to a specific posture angle. At the same time, the technical parameters of each actuator of the working device are obtained, including the crane's lifting height range, slewing angle range, and amplitude range; the winch's maximum wind up / down speed and cable length limit; and the hydraulic arm's joint angle range and end-load capacity.

[0068] The system acquires the current status and position of each actuator, including the current three-dimensional coordinates of the crane hook, the current length of the winch cable, and the current angles of each joint of the hydraulic boom. This status information is collected in real time through devices such as encoders and position sensors. Based on the task execution parameters, the system calculates the target status and position that each actuator needs to achieve. For example, if the task requires lifting cargo from the deck to the hold, the target three-dimensional coordinates of the crane hook need to be calculated; if the task requires adjusting the anchor chain length, the target cable length of the winch needs to be calculated.

[0069] After determining the current and target positions, it is necessary to plan the motion trajectories of each actuator. Motion trajectory planning requires consideration of several constraints: first, kinematic constraints, ensuring the trajectory does not exceed the physical motion range of the actuators; second, dynamic constraints, ensuring acceleration and rate of change of velocity are within the equipment's tolerance range; third, obstacle avoidance constraints, ensuring no collisions with the ship's structure or other equipment occur during motion; and finally, stability constraints, avoiding drastic velocity changes that could cause cargo swaying or equipment vibration.

[0070] For multi-degree-of-freedom actuators, such as hydraulic booms or multi-joint cranes, trajectory planning needs to be performed in joint space or Cartesian space. In joint space planning, an angle change curve from the current angle to the target angle is generated for each joint. In Cartesian space planning, a spatial path from the current position to the target position is generated for the end effector, and the angle sequence of each joint is obtained through inverse kinematics. The trajectory is typically generated using polynomial interpolation or spline curve fitting methods to ensure the continuity of position, velocity, and acceleration.

[0071] When determining the motion sequence, the coordination between various actuators needs to be considered. Some actions must be performed sequentially, such as a crane needing to lift the load to a safe height before it can rotate; others can be performed in parallel, such as adjusting the amplitude while the crane is rotating. By analyzing the dependencies and priority information in the task execution parameters, execution time windows are allocated to the actions of each actuator. For actions requiring coordination, such as multiple winches simultaneously raising and lowering cables to keep the work platform level, synchronization trigger points need to be set in the timing arrangement to ensure that each actuator maintains consistent action at critical moments.

[0072] The planned motion trajectory and motion sequence are encapsulated in a standardized format to form an action sequence instruction. This instruction uses a timestamped action sequence structure, and each action includes information such as actuator identifier, target state parameters, motion speed, acceleration limit, start time, and end time. For actions requiring real-time feedback control, parameters such as position error tolerance and torque limit are also included.

[0073] Power allocation commands and action sequence commands are combined to form a two-layer control command. The combination process requires establishing a timing correlation between the two layers of commands to ensure that the power adjustment of the propulsion system and the execution of the working device's actions are coordinated in time. For example, when the working device initiates a high-power action, the propulsion system needs to increase its power generation in advance to meet the electricity demand; when the ship needs to perform dynamic positioning, the operating speed of the working device needs to be appropriately reduced to minimize disturbance to the ship's attitude. A cross-layer synchronization flag is set in the two-layer control command to identify the control actions that need to be executed collaboratively, and a global timestamp is assigned to each synchronization point, enabling the propulsion system controller and the working device controller to perform collaborative control based on a unified time reference.

[0074] The dual-layer control command also includes an exception handling mechanism, defining strategies for handling abnormal situations such as insufficient power, equipment failure, and sudden environmental changes during execution. For example, if the propulsion system power is insufficient to simultaneously meet the needs of power positioning and operation of the working device, the power of the working device will be reduced or some non-critical actions will be suspended according to preset priorities; if an actuator malfunctions and cannot complete the predetermined action, the backup plan or safety shutdown procedure will be triggered.

[0075] The generated two-tiered control commands are distributed to the lower-level controllers of each hardware system via the ship's internal communication network. Upon receiving the power allocation command, the propulsion system controller adjusts the throttle opening, pitch angle, or motor speed of each propulsion unit to achieve precise power distribution. Upon receiving the action sequence command, the operating device controller drives the hydraulic, motor, or pneumatic systems to control each actuator to complete its actions according to the planned trajectory and timing. Throughout the control process, each controller continuously reports its execution status and feedback data, providing real-time information for subsequent trajectory prediction and correction.

[0076] A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted using this model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control commands are corrected to drive the propulsion system and operating device to collaboratively complete the mode switching, including: Collect attitude change data and propulsion system output data of the hull during the historical operation mode switching process, extract the dynamic response relationship between the two and establish a hull response prediction model. The ship's attitude data and propulsion system control parameters in the dual-layer control command are obtained at the current moment, and input into the ship's response prediction model to predict the ship's motion trajectory in the future time period. The ship's motion trajectory includes the ship's position sequence and attitude angle sequence. Read the boundary coordinates of the safe operating domain, compare the spatial position sequence of the ship with the boundary coordinates, identify the out-of-bounds position points that exceed the boundary, calculate the spatial deviation distance between each out-of-bounds position point and the boundary, and based on the spatial deviation distance and dynamic response relationship, solve in reverse the amount of propulsion system control parameter adjustment required to keep the ship's motion within the safe operating domain as the trajectory correction amount. The trajectory correction is superimposed on the power allocation command for the propulsion system in the dual-layer control command, while the action sequence command for the working device remains unchanged. The corrected power allocation command is sent to the propulsion system, and the action sequence command is sent to the working device.

[0077] During the switching of vessel operation modes, the hull is subject to a combination of factors, including changes in propulsion system output, the movement of operational equipment, and disturbances in the marine environment, resulting in dynamic changes in the hull's attitude and position. To ensure the safety of the switching process, it is necessary to predict the hull's trajectory and correct control commands in real time.

[0078] The system collects attitude change data of the hull during historical operation mode switching processes, including roll, pitch, bow angles, and the hull's position coordinates in the geodetic coordinate system. Simultaneously, it records the output data of the propulsion system at corresponding moments, including parameters such as the rotational speed, thrust magnitude, and thrust direction angle of each propeller. Through analysis of historical data, the dynamic relationship between the propulsion system output and the hull attitude response is extracted. Specifically, the propulsion system output is used as the input variable, and the hull attitude change is used as the output variable. A state-space form hull response prediction model is established using a system identification method. This model can describe the evolution of the hull's attitude and position over time under a given propulsion system control input. The model considers the hull's inertial characteristics, hydrodynamic damping, and the coupling effect between the propellers and the hull, ensuring that the prediction results reflect the hull's true dynamic response characteristics.

[0079] During the mode switching execution phase, real-time attitude data of the hull at the current moment is acquired, including three-axis attitude angles and position coordinates. Power allocation instructions for the propulsion system are extracted from the two-layer control commands. These instructions include the target power, speed setpoint, and thrust vector direction for each thruster. Using the current hull attitude data as the initial state and the propulsion system control parameters as the input sequence for future time periods, these are substituted into the hull response prediction model for rolling prediction. The prediction time domain is set to the future. seconds, prediction step size is The predicted hull trajectory is obtained within seconds. This trajectory includes a sequence of hull positions. and attitude angle sequence ,in For the index of the predicted time, To predict the total number of steps in the time domain, , , They represent the first The predicted position of the ship's hull in the geodetic coordinate system at each moment. , , These represent the roll angle, pitch angle, and bow angle, respectively.

[0080] Read the pre-defined boundary coordinates of the safe operating domain for the current task. The safe operating domain is typically a three-dimensional spatial region enclosed by multiple boundary surfaces, with boundary coordinates given as a set of vertices or boundary equations. Compare the predicted hull position sequence with the spatial positions of the safe operating domain boundaries. For each predicted position point... The distance between a predicted location point and each boundary surface is calculated. If the distance from a predicted location point to a boundary surface is negative or less than a preset safety margin, the location point is determined to be an out-of-bounds location point. After identifying all out-of-bounds location points, the spatial deviation distance between each out-of-bounds location point and its corresponding boundary surface is calculated. This distance reflects the degree to which the ship's trajectory deviates from the safe operating area.

[0081] Based on the spatial deviation distance and the dynamic response relationship established in the hull response prediction model, the required adjustment amounts of propulsion system control parameters to keep the hull motion within the safe operating domain are solved in reverse. Specifically, through model inverse operation or optimization methods, the power adjustment amounts, speed adjustment amounts, and thrust direction adjustment amounts of each propulsion unit are determined, ensuring that the corrected hull trajectory meets the constraints of the safe operating domain. The calculation of the adjustment amounts needs to consider the response speed and output capacity limitations of the propulsion system to ensure that the adjustment commands are within the executable range of the propulsion system. For cases with multiple out-of-bounds positions, the degree of deviation and the time of occurrence of each out-of-bounds point are comprehensively considered, and a unified control parameter adjustment amount is obtained through weighted summation or sequence optimization methods. This adjustment amount is the trajectory correction amount.

[0082] The calculated trajectory correction is superimposed on the power allocation command for the propulsion system within the dual-layer control command. This superposition operation is performed sequentially according to the thruster number, adding the original power allocation value of each thruster to its corresponding adjustment to obtain the corrected power allocation command. During the superposition process, it is necessary to check whether the corrected power value exceeds the thruster's rated power range. If it does, saturation limiting is applied, while maintaining the motion sequence command for the working device unchanged. This ensures the working device executes operations according to the predetermined motion sequence, without altering the operational flow due to the ship's trajectory correction. The corrected power allocation command is sent to the thruster control units of the propulsion system via the ship's control network. The thrusters adjust their speed and thrust output based on the received commands. The motion sequence command is sent to the actuators of the working device through independent control channels, driving the working device to complete predetermined actions such as hoisting, retrieval, and rotation.

[0083] Throughout the mode switching process, the aforementioned prediction and correction procedures are executed cyclically at fixed intervals. Within each control cycle, the current hull attitude data is reacquired, the propulsion system control parameters are updated, and a new round of trajectory prediction and correction calculations is performed. Through rolling prediction and real-time correction, the system can dynamically respond to the impact of marine environmental disturbances and model uncertainties, ensuring that the hull motion remains within the safe operating range. Driven by corrected control commands, the propulsion system and the operating device work collaboratively. The propulsion system compensates for hull position deviations by adjusting thrust output, while the operating device completes the actions required for mode switching according to a predetermined sequence. This coordinated operation achieves safe and efficient mode switching.

[0084] When the predicted trajectory indicates no risk of exceeding the limits in the future time domain, the trajectory correction is zero, the dual-layer control commands remain unchanged, and the propulsion system and operating devices execute according to the initially planned switching strategy. When a risk of exceeding the limits is detected, the trajectory correction is dynamically adjusted based on the degree of deviation; the greater the deviation, the greater the correction, thus achieving rapid correction of the ship's motion. The magnitude of the correction is also constrained by the propulsion system's response characteristics to prevent oscillations or saturation in the propulsion system output due to excessively rapid changes in control commands.

[0085] In certain complex operational scenarios, the boundaries of the safe operational domain may change dynamically over time, such as in multi-vehicle collaborative operations or when dynamic obstacles are present. In such cases, the boundary coordinates of the safe operational domain need to be updated in real time, and the boundary data used in the prediction and correction process must be the latest boundary information at the current moment. By reading the updated boundary coordinates in real time, it is ensured that trajectory comparison and correction calculations are always based on accurate safety constraints.

[0086] The accuracy of the ship's response prediction model directly affects the effectiveness of trajectory prediction and correction. To improve model accuracy, actual ship motion data can be continuously collected and compared with predicted data during mode switching, and online learning or adaptive algorithms can be used to correct model parameters in real time. Through continuous model optimization, the deviation between the predicted trajectory and the actual trajectory gradually decreases, improving the robustness and adaptability of the control system.

[0087] Obtain the current hull attitude data and propulsion system control parameters from the two-layer control commands, input them into the hull response prediction model, and predict the hull trajectory over a future time period, including: Set a rolling prediction time window and a window update cycle. During mode switching, the prediction task is triggered cyclically according to the window update cycle. When each prediction task is triggered, the ship's position coordinates, the ship's three-axis attitude angles and the ship's six-degree-of-freedom velocity components are collected and the collected data are fused into the ship's attitude state. The power allocation command for the propulsion system is parsed from the two-layer control command, and the allocated power value and thrust vector direction of each propulsion unit are extracted. Based on the thrust characteristic curve of each propulsion unit, the allocated power value is converted into a thrust output value, and the thrust output value and thrust vector direction are combined into propulsion system control parameters. The hull attitude state is used as the initial state input of the hull response prediction model, and the propulsion system control parameters are used as the control input to drive the hull response prediction model to perform dynamic calculations. Within the rolling prediction time window, iterative calculations are performed according to a preset simulation step size. In each iteration, the resultant force and resultant torque on the hull are calculated based on the propulsion system control parameters, and the hull position and attitude are updated. The hull position and attitude output by each iteration step are arranged in a time sequence to generate the hull motion trajectory.

[0088] During mode switching, to ensure the ship's motion remains within a controllable range, real-time prediction of the ship's trajectory over future time periods is necessary. The core of the rolling prediction mechanism lies in establishing a dynamically updated time window, the length of which is set based on the ship's response characteristics and the complexity of the operating environment. For large engineering vessels, the rolling prediction time window is typically set to 30 to 120 seconds, while the window update cycle is determined based on the control system's computing power and real-time requirements, generally ranging from 1 to 5 seconds. The selection of the window update cycle requires a balance between prediction accuracy and computational load; an excessively short update cycle leads to excessive computational resource consumption, while an excessively long update cycle causes the prediction results to lag behind actual changes in the state.

[0089] At the end of each window update cycle, the prediction task is automatically triggered. At this time, the ship's current state information is collected through the shipborne sensor system. The ship's position coordinates are obtained through a differential global positioning system, with positioning accuracy down to the centimeter level. This includes the longitude, latitude, and altitude of the ship's center of mass in the geodetic coordinate system. After coordinate transformation, the three-dimensional position coordinates of the ship in the local operating coordinate system are obtained. The ship's three-axis attitude angles are measured by an inertial measurement unit, including roll, pitch, and bow angles, with a measurement accuracy typically better than 0.1 degrees. The ship's six-degree-of-freedom velocity components include longitudinal velocity, lateral velocity, vertical velocity, and angular velocity about the three coordinate axes. These velocity components are obtained by numerical differentiation of the position and attitude data or by direct measurement through velocity sensors.

[0090] The collected position coordinates, attitude angles, and velocity components are fused to form a complete ship attitude state. The data fusion process employs a Kalman filter algorithm to weight and fuse measurement data from different sensors, eliminating measurement noise and sensor drift. The fused ship attitude state is represented as a state vector containing 12 components, corresponding to the ship's position, attitude, and velocity information. The construction of the state vector follows standard representation methods in ship kinematics to ensure correct parsing and processing by subsequent prediction models.

[0091] When parsing propulsion system control information from two-layer control commands, it is necessary to identify the data structure and encoding format of the commands. Power allocation commands are typically transmitted in the form of structured data packets, which contain the propulsion unit's identification code, allocated power value, and thrust vector direction. The propulsion unit identification code is used to distinguish different propellers, such as main thrusters, side thrusters, and bow thrusters. The allocated power value is in kilowatts and represents the power level that the propulsion unit should output in the current control cycle. The thrust vector direction is described by azimuth and pitch angles. The azimuth angle is defined as the angle between the thrust direction and the ship's longitudinal axis in the horizontal plane, and the pitch angle is defined as the angle of inclination of the thrust direction relative to the horizontal plane.

[0092] The thrust characteristic curves of each propulsion unit describe the nonlinear relationship between power input and thrust output. These curves are typically obtained through bench tests or computational fluid dynamics simulations of the propulsion unit and stored in the control system as lookup tables or polynomial fitting functions. When converting the allocated power value into a thrust output value, the corresponding thrust characteristic curve is retrieved based on the propulsion unit's identification code. The allocated power value is then substituted into the curve function for interpolation calculation to obtain the thrust output value at that power level. The unit of thrust output value is kilonewtons (kN), representing the actual thrust that the propulsion unit can generate under the current operating conditions.

[0093] The thrust output value and thrust vector direction are combined to form the propulsion system control parameters. For ships equipped with multiple propulsion units, the thrust output value and thrust vector direction of each propulsion unit need to be organized into vector form. The dimension of the thrust output value vector is equal to the number of propulsion units, and the thrust vector direction is represented in the form of azimuth and pitch vectors. These parameters together constitute the control input of the propulsion system in the prediction time domain, providing driving force information for the hull response prediction model.

[0094] The ship response prediction model is based on the six-degree-of-freedom dynamic equations of ship motion. This model considers the forces acting on the hull under the influence of waves, wind, and ocean currents, as well as the thrust and torque generated by the propulsion system. The model's state variables are the aforementioned hull attitude states, and the control inputs are the propulsion system control parameters. During model initialization, the fused hull attitude state at the current moment is assigned to the model's initial state variables to ensure that the prediction starting point is consistent with the actual state. The propulsion system control parameters are input into the model's control input port, and the model calculates the resultant force and resultant torque generated by the propulsion system on the hull based on these inputs.

[0095] The dynamic calculation process employs numerical integration to solve the ship's motion equations. Within the rolling prediction time window, the model is discretized according to a preset simulation step size. The simulation step size must satisfy numerical stability conditions and is typically between 0.1 and 0.5 seconds. An excessively large simulation step size leads to accumulated numerical errors, affecting prediction accuracy; conversely, an excessively small step size increases computational load and reduces prediction efficiency. Within each simulation step, the model performs one iterative calculation.

[0096] The first step in the iterative calculation is to calculate the propulsion force and propulsion torque acting on the hull based on the propulsion system control parameters. The propulsion force is obtained by vector decomposition and superposition of the thrust output values ​​of each propulsion unit according to the thrust vector direction. The decomposition process needs to consider the installation position and orientation of the propellers in the hull coordinate system. The propulsion torque is generated by the lever arm of the propulsion force relative to the hull's center of mass. The calculation of the lever arm depends on the spatial arrangement of the propellers. In addition to the propulsion force, the model also needs to calculate the marine environmental loads, including wave forces, wind forces, and ocean current forces. These environmental loads are calculated based on current marine environmental data and the hull's hydrodynamic coefficients.

[0097] The propulsion force, propulsion torque, and environmental loads are vector-superimposed to obtain the resultant force and resultant torque acting on the hull. The resultant force acts on the hull's center of mass, causing linear acceleration; the resultant torque acts on the hull, causing angular acceleration. Based on Newton's second law and Euler's equations, the linear and angular accelerations of the hull are calculated using the hull's mass, moment of inertia, and additional mass and moment of inertia. The acceleration is integrated over time to update the hull's velocity components; the velocity components are then integrated over time to update the hull's position coordinates and attitude angles. The integration process employs the fourth-order Runge-Kutta method or other high-precision numerical integration algorithms to ensure the accuracy of the calculation results.

[0098] After each iteration, the updated hull position and attitude are used as the initial state for the next iteration, and the calculation of the next simulation step continues. This iterative process continues until the entire rolling prediction time window is covered. During the iteration, the hull position and attitude data output at each simulation step are recorded and arranged in chronological order to form a continuous trajectory curve. This trajectory curve is the predicted hull motion trajectory, containing information on the hull's position and attitude changes over future time periods.

[0099] The generation of predicted trajectories not only provides fundamental data for subsequent safety domain boundary comparisons but also reveals the dynamic characteristics of hull motion during mode switching. By analyzing the curvature, rate of change of velocity, and rate of change of attitude of the predicted trajectory, potential unstable motion modes or excessive attitude deviations can be identified. This information provides a basis for trajectory correction and control command optimization, ensuring that the vessel maintains a safe and stable operating state throughout mode switching. The advantage of the rolling prediction mechanism lies in its ability to dynamically adjust prediction results based on changes in the actual state. In each window update cycle, the prediction task restarts, using the latest hull attitude state and control commands for prediction, thereby eliminating the cumulative effects of model errors and external disturbances. This closed-loop prediction method significantly improves the reliability and practicality of predictions, providing solid technical support for the intelligent switching of vessel operation modes.

[0100] A second aspect of the present invention provides an intelligent switching and control system for ship operation modes, comprising: The data acquisition unit is used to collect data from the ship's onboard sensors, including hull attitude data, propulsion system status data, operating device operation data, and marine environment data. The state modeling unit is used to establish a multi-dimensional state space model that integrates the semantics of the operation task and the physical constraints of the ship hull. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector, and the current operation mode identifier is identified by combining a predefined pattern cluster. The optimization decision unit is used to establish a comprehensive optimization objective function that includes time cost and energy consumption based on the preset operation process and the current operation mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the operation device, and solve for the target operation mode and hierarchical switching strategy. The instruction generation unit is used to generate two-layer control instructions based on the hierarchical switching strategy. The two-layer control instructions include power allocation instructions for the propulsion system and action sequence instructions for the working device. The prediction and correction unit is used to establish a hull response prediction model, and to perform rolling prediction of the hull motion trajectory during the switching process using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operation domain, the trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and the operation device to complete the mode switching in a coordinated manner.

[0101] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0102] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0103] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent switching and control of ship operation modes, characterized in that, include: Collect shipboard sensor data from the target vessel, including hull attitude data, propulsion system status data, operational equipment data, and marine environment data; A multi-dimensional state space model integrating the semantics of the operation task and the physical constraints of the ship is established. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector. The current operation mode identifier is identified by combining the predefined pattern cluster. Based on the preset work process and the current work mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the work device, a comprehensive optimization objective function including time cost and energy consumption is established, and the target work mode and hierarchical switching strategy are obtained by solving it. Based on the hierarchical switching strategy, a two-level control command is generated, which includes a power allocation command for the propulsion system and an action sequence command for the working device. A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted by rolling using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and operating device to complete the mode switching in a coordinated manner.

2. The method of claim 1, wherein, A multi-dimensional state-space model integrating operational task semantics and ship hull physical constraints is established. The shipborne sensor data is projected and transformed within this model to obtain a state vector. This vector is then combined with a predefined pattern cluster to identify the current operational mode, including: Extract historical sensor data and corresponding operation mode labels from multiple operation processes from the historical operation database; Feature extraction is performed on the historical sensor data to obtain a set of task semantic features and a set of physical constraint features, and statistical distribution features are extracted to determine the dimensional structure of the state space and the value range of each dimension, and a multidimensional state space model is established. In the multidimensional state space model, the historical sensor data is grouped and projected according to the operation mode label, and the projection results of each group are clustered to form a predefined mode cluster. The predefined mode cluster includes the center coordinates and distribution boundary of each mode cluster. A projection transformation mapping function is constructed from the original sensor data space to the multidimensional state space model. The projection transformation mapping function achieves dimensional alignment by maintaining the correlation between features. The shipborne sensor data is input into the projection transformation mapping function to obtain the corresponding state vector in the multidimensional state space model. Calculate the spatial distance between the state vector and the center coordinates of each pattern cluster in the predefined pattern cluster, and select the pattern type corresponding to the pattern cluster with the smallest spatial distance as the current operation mode identifier.

3. The method of claim 2, wherein, Feature extraction is performed on the historical sensor data to obtain a task semantic feature set and a physical constraint feature set, and statistical distribution features are extracted. The dimensional structure of the state space and the value range of each dimension are determined, and a multidimensional state space model is established, including: Based on the historical sensor data, historical operation data of the working device and historical propulsion system status data are extracted. The historical operation data of the working device is segmented in time sequence to extract the device action mode. The historical propulsion system status data is analyzed by power spectrum to extract the power output frequency domain features. The device action mode and power output frequency domain features are combined to form a task semantic feature set. Historical hull attitude data and historical marine environment data are extracted from the historical sensor data. The response correlation function between hull attitude and environmental disturbance is calculated to obtain the hull attitude offset amplitude and recovery time constant corresponding to different environmental disturbance intensities. Based on the hull stability criterion, the boundary values ​​of the attitude offset amplitude and recovery time constant are determined, and the boundary values ​​constitute a set of physical constraint features. Correlation analysis is performed on the task semantic feature set to identify the dominant features and the number of the dominant features is determined as the number of task dimensions. Extreme value statistics and distribution fitting are performed on the physical constraint feature set to determine the number of physical dimensions and boundary range. Orthogonal coordinate axes are set and discrete grids are divided according to the numerical range of the dominant feature and the boundary range. The structure and range of the discrete grids are used as the dimensional structure and value range of each dimension of the multidimensional state space model.

4. The method of claim 1, wherein, Based on the preset work process and the current work mode identifier, and combining the ship's dynamic response characteristics and the coupling relationship of the work equipment, a comprehensive optimization objective function including time cost and energy consumption is established. Solving this function yields the target work mode and hierarchical switching strategy, including: Extract the job stage transfer rules from the preset job flow, and find the set of reachable job modes based on the current job mode identifier; Collect fuel consumption rate data and response lag data of the ship's power system under different loads, collect drive power data and switching preparation time data of the working device under different working conditions, and calculate the energy consumption rate per unit time and the time window required for mode switching. For each work mode in the set of reachable work modes, obtain the corresponding action instruction sequence, query the drive power data according to the action instruction sequence, calculate the total energy consumption in combination with the energy consumption rate per unit time, and calculate the total work time according to the execution duration of the action instruction sequence and the time window required for the mode switching. Multiply the total operation time by the time cost coefficient, multiply the total energy consumption by the energy cost coefficient, and add them together to obtain the comprehensive optimization objective function value; The operation mode corresponding to the minimum comprehensive optimization objective function value is selected as the target operation mode. The control parameters that need to be adjusted from the current operation mode to the target operation mode are identified, and a hierarchical switching strategy is generated according to the degree of influence on hull stability.

5. The method of claim 1, wherein, Based on the hierarchical switching strategy, a two-layer control command is generated. This two-layer control command includes power allocation commands for the propulsion system and action sequence commands for the operating device, including: The hierarchical switching strategy is analyzed to obtain the control hierarchy division result. The control hierarchy involved in the propulsion system in the control hierarchy division result is merged into the propulsion system control layer, and the control hierarchy involved in the working device is merged into the working device control layer. Extract the power demand parameters corresponding to the control layer of the propulsion system, obtain the rated power and current available power of each propulsion unit of the propulsion system, calculate the power gap between the power demand parameters and the current available power of each propulsion unit, generate the power adjustment amount of each propulsion unit based on the power gap and the load adjustment rate of each propulsion unit, and encapsulate the power adjustment amount into a power allocation instruction. Extract the task execution parameters corresponding to the control layer of the working device, obtain the motion capability range and current state position of each actuator of the working device, calculate the target state position to be reached by each actuator based on the task execution parameters, determine the motion trajectory and motion sequence of each actuator based on the current state position and the target state position, encapsulate the motion trajectory and motion sequence into action sequence instructions, and combine them with the power allocation instructions to form a two-layer control instruction.

6. The method of claim 1, wherein, A hull response prediction model is established, and the hull motion trajectory during the switching process is predicted using this model. By comparing the predicted trajectory with the boundary of the safe operating domain, a trajectory correction amount is generated and the dual-layer control commands are corrected to drive the propulsion system and operating device to collaboratively complete the mode switching, including: Collect attitude change data and propulsion system output data of the hull during the historical operation mode switching process, extract the dynamic response relationship between the two and establish a hull response prediction model. The ship's attitude data and propulsion system control parameters in the dual-layer control command are obtained at the current moment, and input into the ship's response prediction model to predict the ship's motion trajectory in the future time period. The ship's motion trajectory includes the ship's position sequence and attitude angle sequence. Read the boundary coordinates of the safe operating domain, compare the spatial position sequence of the ship with the boundary coordinates, identify the out-of-bounds position points that exceed the boundary, calculate the spatial deviation distance between each out-of-bounds position point and the boundary, and based on the spatial deviation distance and dynamic response relationship, solve in reverse the amount of propulsion system control parameter adjustment required to keep the ship's motion within the safe operating domain as the trajectory correction amount. The trajectory correction is superimposed on the power allocation command for the propulsion system in the dual-layer control command, while the action sequence command for the working device remains unchanged. The corrected power allocation command is sent to the propulsion system, and the action sequence command is sent to the working device.

7. The method of claim 6, wherein, Obtain the current hull attitude data and propulsion system control parameters from the two-layer control commands, input them into the hull response prediction model, and predict the hull trajectory over a future time period, including: Set a rolling prediction time window and a window update cycle. During mode switching, the prediction task is triggered cyclically according to the window update cycle. When each prediction task is triggered, the ship's position coordinates, the ship's three-axis attitude angles and the ship's six-degree-of-freedom velocity components are collected and the collected data are fused into the ship's attitude state. The power allocation command for the propulsion system is parsed from the two-layer control command, and the allocated power value and thrust vector direction of each propulsion unit are extracted. Based on the thrust characteristic curve of each propulsion unit, the allocated power value is converted into a thrust output value, and the thrust output value and thrust vector direction are combined into propulsion system control parameters. The hull attitude state is used as the initial state input of the hull response prediction model, and the propulsion system control parameters are used as the control input to drive the hull response prediction model to perform dynamic calculations. Within the rolling prediction time window, iterative calculations are performed according to a preset simulation step size. In each iteration, the resultant force and resultant torque on the hull are calculated based on the propulsion system control parameters, and the hull position and attitude are updated. The hull position and attitude output by each iteration step are arranged in a time sequence to generate the hull motion trajectory.

8. An intelligent switching and control system for ship operation modes for implementing the method according to any one of claims 1 - 7, characterized in that, include: The data acquisition unit is used to collect data from the ship's onboard sensors, including hull attitude data, propulsion system status data, operating device operation data, and marine environment data. The state modeling unit is used to establish a multi-dimensional state space model that integrates the semantics of the operation task and the physical constraints of the ship hull. The shipborne sensor data is projected and transformed in the multi-dimensional state space model to obtain a state vector, and the current operation mode identifier is identified by combining a predefined pattern cluster. The optimization decision unit is used to establish a comprehensive optimization objective function that includes time cost and energy consumption based on the preset operation process and the current operation mode identifier, combined with the ship's dynamic response characteristics and the coupling relationship of the operation device, and solve for the target operation mode and hierarchical switching strategy. The instruction generation unit is used to generate two-layer control instructions based on the hierarchical switching strategy. The two-layer control instructions include power allocation instructions for the propulsion system and action sequence instructions for the working device. The prediction and correction unit is used to establish a hull response prediction model, and to perform rolling prediction of the hull motion trajectory during the switching process using the hull response prediction model. By comparing the predicted trajectory with the boundary of the safe operation domain, the trajectory correction amount is generated and the dual-layer control command is corrected to drive the propulsion system and the operation device to complete the mode switching in a coordinated manner.

9. An electronic device, comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images