Intelligent auxiliary decision-making method for ship entering and leaving port and berthing and leaving berthing based on multi-source data

Abstract

The embodiment of the invention relates to the technical field of ship scheduling, in particular to a multi-source data-based intelligent auxiliary decision-making method for ship entering and leaving a port and berthing and leaving the port, and the real-time performance and accuracy of port scheduling decision-making can be improved by fusing multi-source heterogeneous data to construct a ship navigation situation awareness system. The space-time alignment technology effectively eliminates space-time dislocation of environment interference factors and enhances the characterization capability of the ship motion state and environment coupling relation. Quantitative evaluation of multi-dimensional navigation risks is realized based on a ship behavior prediction model, and potential risks such as channel deviation and berthing conflicts can be identified. A collaborative decision-making scheme considering safety and efficiency is generated through a collaborative optimization mechanism of the dynamic scheduling rule base and the risk indexes, and ship trajectory control precision is optimized through dynamic planning of navigational speed adjustment and steering time sequence. Closed-loop linkage of port resource allocation and ship navigation control is achieved through automatic instruction generation and execution, and the reliability and the response speed of the port operation process can be ensured.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of ship scheduling, and in particular to an intelligent auxiliary decision-making method for ship entry, exit, berthing and unberthing based on multi-source data. Background Art

[0002] Traditional port dispatching systems have defects such as low multi-source data integration, delayed risk prediction, and insufficient decision-making automation. Existing technologies lack temporal and spatial correlation analysis between dynamic environmental data such as meteorology and hydrology and ship trajectories, resulting in deviations in the assessment of the impact of environmental interference on navigation safety. Ship behavior prediction mostly relies on single-dimensional parameters, making it difficult to accurately quantify the compound risk of channel deviation and berthing conflict. The static dispatching rule base cannot adapt to the dynamic changes in the occupation status of port resources, and the coordination between berth allocation and speed control strategies is poor. Frequent manual intervention leads to delayed emergency response, and the real-time nature of ship trajectory adjustment and port equipment linkage is limited. In addition, there is a lack of effective integration of historical navigation data and real-time decision-making processes, which restricts the ability to continuously optimize dispatching strategies. The above problems have resulted in low port operation efficiency and high risks of safety accidents. It is urgent to build a data-driven intelligent decision-making system to achieve a systematic breakthrough. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, one of the purposes of an embodiment of the present invention is to provide an intelligent auxiliary decision-making method for ship entry, exit and berthing based on multi-source data.

[0004] The embodiment of the present invention provides an intelligent auxiliary decision-making method for ship entry and exit and berthing and unberthing based on multi-source data, including: Acquire multi-source heterogeneous data in the target port area, wherein the multi-source heterogeneous data includes real-time meteorological data, hydrological monitoring data, ship dynamic trajectory data, port facility status data and historical navigation record data; Performing spatiotemporal alignment and feature fusion on the multi-source heterogeneous data to generate a navigation situation feature set of the target ship, wherein the navigation situation feature set includes ship navigation state parameters, a set of environmental interference factors, and a sequence of port resource occupancy states; Input the navigation situation feature set into a pre-trained ship behavior prediction model, and output a navigation risk index set of the target ship under multiple decision dimensions, wherein the navigation risk index set includes a channel deviation probability, a berthing conflict probability, a navigation path feasibility score, and a port operation efficiency impact coefficient; Based on the navigation risk indicator set and the preset port dynamic scheduling rule base, a collaborative decision-making scheme for the entry, exit and berthing of the target ship is generated, wherein the collaborative decision-making scheme includes a speed adjustment sequence, a turn timing set, a berth allocation priority and a berthing and unberthing operation time window; A ship control instruction set is generated according to the collaborative decision-making scheme, and the ship control instruction set is sent to the navigation equipment and port operation equipment of the target ship through the port dispatching system to dynamically adjust the ship's navigation trajectory and port resource allocation status.

[0005] In one technical solution, the step of performing spatiotemporal alignment and feature fusion on the multi-source heterogeneous data to generate a navigation situation feature set of the target ship includes: Extracting the ship position sequence, speed sequence and heading angle sequence from the ship dynamic trajectory data to construct a ship kinematics feature vector; Performing spatial grid matching on the wind speed, wind direction, and visibility in the real-time meteorological data and the tidal height and flow velocity in the hydrological monitoring data to generate an environmental interference factor matrix; Constructing a time series of port resource occupancy status according to the berth occupancy status, loading and unloading equipment working status and waterway congestion index in the port facility status data; The ship kinematic feature vector, environmental interference factor matrix and port resource occupancy state sequence are time-stamp synchronized and spatial coordinate mapped to generate a multi-dimensional fusion navigation situation feature set, in which each feature unit is associated with the real-time position and future predicted time point of the target ship.

[0006] In one technical solution, the training process of the ship behavior prediction model includes: Collect historical multi-source heterogeneous data and corresponding actual ship navigation decision records to construct a training sample set, which includes a historical navigation situation feature set and annotated navigation risk indicator true values; Construct a multimodal deep learning network, which includes a spatiotemporal convolution module, an attention weight allocation module, and a dynamic feature interaction module, wherein: The spatiotemporal convolution module is used to extract the spatial correlation between the local motion pattern of the ship trajectory and the environmental interference factors; The attention weight allocation module dynamically adjusts the contribution of meteorological data, hydrological data and port status data to the impact of ship behavior; The dynamic feature interaction module simulates the coupling effect between the ship motion characteristics and the environmental interference factors at continuous time steps; The loss function of the multimodal deep learning network is optimized by a back-propagation algorithm, wherein the loss function includes the mean square error of the channel deviation probability, the cross entropy of the berthing conflict probability, and the ranking loss of the navigation path feasibility score.

[0007] In one technical solution, the generating of a collaborative decision-making scheme for the entry, exit and berthing of the target ship includes: Determine the channel right of way level of the target ship according to the channel deviation probability and the channel priority table in the port dynamic scheduling rule base; Based on the berthing conflict probability and the berth allocation history data, calculating the berth suitability score of the target ship, the score including the matching degree between the ship draft and the berth water depth, and the ratio of the ship length to the length of the berth free area; Combining the navigation path feasibility score with the port operation efficiency impact coefficient, generating a set of constraint conditions for the speed adjustment sequence, wherein the constraint conditions include a minimum safe speed threshold, a maximum turning angle rate, and an environmental interference tolerance; The waterway right level, berth suitability score and speed adjustment constraints are weightedly integrated through a multi-objective optimization algorithm to generate a collaborative decision-making plan that includes time window constraints and resource conflict avoidance.

[0008] In one technical solution, the generation and execution of the ship control instruction set includes: Decomposing the speed adjustment sequence in the collaborative decision-making scheme into discrete control instructions, each instruction including a target speed value, an acceleration limit and an effective time interval; generating a steering angle control signal sequence according to the steering opportunity set, wherein the signal sequence is matched with the terrain feature points of the port channel in real time; Mapping the berth allocation priority into a berth guidance instruction, wherein the instruction includes berth coordinates, berthing angle, and activation timing of cable mooring equipment; The ship control instruction set is pushed to the target ship's autopilot system and port operation control terminal in real time through the port dispatching system, and the ship's response status and port equipment feedback data during the execution of the instructions are monitored.

[0009] In one technical solution, the method further includes: Real-time collection of the execution feedback data of the ship control instruction set, including the actual speed deviation of the ship, the steering angle execution error and the change in berth occupancy status; Inputting the execution feedback data into the ship behavior prediction model for online incremental learning, and dynamically adjusting the environmental factor contribution parameters in the attention weight allocation module of the model; Recalculate the navigation risk indicator set based on the updated ship behavior prediction model to generate a revised collaborative decision-making plan; When it is detected that the waterway congestion index exceeds a preset threshold or the probability of berth conflict continues to rise, the real-time reconstruction process of the collaborative decision-making scheme is triggered, and a collaborative collision avoidance instruction is sent to the associated ships.

[0010] In one technical solution, the online incremental learning process includes: Extracting the ship response delay time, environmental interference mutation events and equipment failure abnormality codes from the execution feedback data to construct a model error feature vector; Through the sliding time window mechanism, the deviation between historical prediction results and actual navigation data is statistically analyzed to identify the weak links in the stability of model prediction; The knowledge distillation technology is used to take the fully trained ship behavior prediction model as the teacher model to generate the distillation loss function of the lightweight student model. The student model is fine-tuned using the execution feedback data and the distillation loss function, and the fine-tuned model parameters are synchronized to the edge computing node of the port scheduling system.

[0011] In one technical solution, the generation and distribution of the collaborative collision avoidance instruction includes: When it is detected that the predicted navigation paths of multiple ships have overlapping areas in the time and space dimensions, the center coordinates of the overlapping areas and the conflict time windows are extracted; Calculate the collision avoidance priority weight of each ship, the weight being based on the ship's tonnage, the cargo hazard level, and the current adjustable speed range; Generate differentiated speed adjustment suggestions and steering angle compensation amounts according to the collision avoidance priority weights, and establish a communication negotiation protocol between ships; The collision avoidance instructions and negotiation protocols are broadcast to associated ships through the ship self-organizing network, and instruction confirmation signals from each ship are received to update the collaborative decision-making plan.

[0012] In one technical solution, the method further includes: During the berthing and unberthing process of the target ship, real-time monitoring of cable tension sensor data, tidal change rate and workload of terminal loading and unloading equipment; When it is detected that the cable tension exceeds the material strength threshold or the tidal change causes insufficient berth water depth, an emergency unberthing instruction set is generated, which includes emergency thruster start-up parameters, cable quick release sequence and port tugboat dispatch priority; The current berthing operation is forcibly interrupted through the port emergency control system, and the automated execution process of the emergency unberthing instruction set is activated.

[0013] In one technical solution, the method further includes: After the ship completes the port entry and exit and berthing and unberthing operations, the decision execution data and external environment monitoring data of the whole process are collected to build a case knowledge graph; Extract key decision nodes, risk handling modes and resource scheduling optimization paths in the case knowledge graph, generate empirical rules and supplement them to the port dynamic scheduling rule base; Simulating decision paths under different environmental interference scenarios through a reinforcement learning algorithm, and updating the weight allocation strategy of the multi-objective optimization algorithm; The optimized decision logic is deployed to the digital twin platform of the port dispatching system to achieve closed-loop verification of historical case data and real-time decision-making processes.

[0014] An embodiment of the present invention also provides an intelligent auxiliary decision-making system, including a processor and a memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the above-mentioned intelligent auxiliary decision-making method for ship entry and exit and berthing based on multi-source data.

[0015] An embodiment of the present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned intelligent auxiliary decision-making method for ship entry, exit and berthing based on multi-source data is implemented.

[0016] The embodiment of the present invention constructs a ship navigation situation awareness system by integrating multi-source heterogeneous data, significantly improving the real-time and accuracy of port scheduling decisions. The spatiotemporal alignment technology of real-time meteorological and hydrological data effectively eliminates the spatiotemporal dislocation of environmental interference factors and enhances the characterization ability of the ship's motion state and the coupling relationship between the environment. The ship behavior prediction model based on deep learning realizes the quantitative assessment of multi-dimensional navigation risks and accurately identifies potential risks such as channel deviation and berthing conflicts. The collaborative optimization mechanism of the dynamic scheduling rule base and risk indicators generates a collaborative decision-making plan that takes into account both safety and efficiency, and optimizes the ship's trajectory control accuracy through dynamic planning of speed adjustment and steering timing. The automated instruction generation and execution system realizes the closed-loop linkage of port resource allocation and ship navigation control, ensures the reliability and response speed of port operation processes in complex operating environments, and improves the overall port navigation efficiency and safety level. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of an intelligent auxiliary decision-making method for ship entry, exit, berthing and unberthing based on multi-source data provided by an embodiment of the present invention.

[0019] Figure 2 A block diagram of an intelligent decision-making support system provided by an embodiment of the present invention.

[0020] icon: 100-Intelligent decision-making support system; 101 - processor; 102 - memory; 103 - bus. DETAILED DESCRIPTION

[0021] The exemplary embodiments disclosed in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the embodiments of the present invention are shown in the accompanying drawings, it should be understood that the embodiments of the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the embodiments of the present invention and to fully convey the scope of the embodiments of the present invention to those skilled in the art.

[0022] In order to better understand the above technical scheme, the technical scheme of the embodiment of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiment of the present invention and the specific features in the embodiments are detailed descriptions of the technical scheme of the embodiment of the present invention, rather than limitations on the technical scheme of the embodiment of the present invention. In the absence of conflict, the embodiment of the present invention and the technical features in the embodiments can be combined with each other.

[0023] Figure 1 A flowchart of an intelligent auxiliary decision-making method for ship entry, exit, berthing and unberthing based on multi-source data provided according to an embodiment of the present invention, which is applied to an intelligent auxiliary decision-making system, includes steps 110 to 150.

[0024] Step 110: Acquire multi-source heterogeneous data in the target port area, wherein the multi-source heterogeneous data includes real-time meteorological data, hydrological monitoring data, ship dynamic trajectory data, port facility status data, and historical navigation record data.

[0025] Step 120: Performing spatiotemporal alignment and feature fusion on the multi-source heterogeneous data to generate a navigation status feature set of the target ship, wherein the navigation status feature set includes ship navigation status parameters, a set of environmental interference factors, and a port resource occupancy status sequence.

[0026] Step 130: Input the navigation situation feature set into a pre-trained ship behavior prediction model, and output a set of navigation risk indicators of the target ship under multiple decision dimensions, wherein the set of navigation risk indicators includes channel deviation probability, berthing conflict probability, navigation path feasibility score and port operation efficiency impact coefficient.

[0027] Step 140: Based on the navigation risk indicator set and the preset port dynamic scheduling rule base, a collaborative decision-making plan for the entry and exit and berthing and unberthing of the target ship is generated, and the collaborative decision-making plan includes a speed adjustment sequence, a turning timing set, a berth allocation priority and a berthing and unberthing operation time window.

[0028] Step 150: Generate a ship control instruction set according to the collaborative decision-making scheme, and send the ship control instruction set to the navigation equipment and port operation equipment of the target ship through the port dispatching system to dynamically adjust the ship's navigation trajectory and port resource allocation status.

[0029] It can be understood that based on the above steps 110 to 150, the intelligent decision-making support system obtains real-time meteorological data, hydrological monitoring data, ship dynamic trajectory data, port facility status data and historical navigation record data in the target port area through a multi-source heterogeneous data acquisition module. Real-time meteorological data includes parameters such as wind speed, wind direction, visibility, and precipitation intensity, among which the wind speed measurement accuracy reaches 0.1 meters per second, and the data update frequency is once a minute; hydrological monitoring data covers indicators such as tidal height, water flow speed, and wave period, and the data is uploaded every 30 seconds through the buoy sensor network deployed in the port waters. The ship's dynamic trajectory data is collected through the fusion of the ship's automatic identification system (Automatic Identification System, AIS) and radar, including dynamic parameters such as ship latitude and longitude coordinates, speed, heading angle, draft depth, etc., and the trajectory sampling interval is 10 seconds. The port facility status data involves the operating status of the dock crane, the berth occupancy status, and the mooring pile load parameters, which are transmitted in real time through the industrial Internet of Things sensors, and the data accuracy reaches the millimeter level. Historical navigation record data is stored in the port data center, including all ship entry and exit times, route planning records, accident reports and operation logs in the past five years. The total amount of data exceeds 500TB and is managed using a distributed storage architecture.

[0030] In the stage of spatiotemporal alignment and feature fusion, the intelligent decision-making system normalizes and timestamp-aligns multi-source heterogeneous data through the data preprocessing module. Specifically, the system uses a sliding time window algorithm to unify data streams of different frequencies to a 5-second time granularity. For example, the meteorological data updated every minute is decomposed into a second-level data stream through a linear interpolation algorithm, and the 30-second period tidal data is synchronized to a 5-second timestamp through a cubic spline interpolation algorithm. The spatial alignment module maps the ship's dynamic trajectory data to a polar coordinate system with the dock as the origin based on the port geographic information system (GIS). The coordinate system division accuracy is 0.001 radians, realizing the spatial association between the ship's position and port facilities. The feature fusion engine uses the multi-head attention mechanism in the deep neural network to jointly encode the ship's navigation status parameters (speed, heading angle), environmental interference factors (wind speed, water flow vector) and port resource occupancy status (berth idle time, crane operation queue) to generate a 768-dimensional navigation situation feature vector. The feature vector is modeled through a gated recurrent unit (GRU) network to capture the dynamic coupling relationship between the ship's motion trend and environmental impact, and finally generates a navigation status feature set of the target ship. The feature dimension is extended to 1024 dimensions, and the time coverage range is the next 30-minute prediction window.

[0031] The ship behavior prediction model adopts an integrated learning framework, including three sub-models: Gradient Boosting Decision Tree (GBDT), Long Short-Term Memory (LSTM) and Random Forest. After the input layer receives the 1024-dimensional vector of the navigation situation feature set, the GBDT model first calculates the probability of channel deviation, which is generated based on 20,000 positive samples and 150,000 negative samples in historical accident data. Feature importance analysis shows that the weights of heading angle variance and wind speed mutation value account for 32.7% and 28.5% respectively. The LSTM network processes the port resource occupancy state sequence, captures the berth allocation time series pattern through 256 hidden units, and outputs the predicted value of berthing conflict probability, with a mean squared error (MSE) of 0.017 on the validation set. The random forest model is used to model the score of the feasibility of the navigation path, using 500 decision trees for integration. The input features include the difference between the ship's turning radius, the channel water depth and the real-time draft. The model output score ranges from 0 to 100 points, with a resolution of 0.1 points. Finally, the model fusion layer integrates the output results of the three sub-models into a set of navigation risk indicators through a weighted voting mechanism. The port operation efficiency impact coefficient is calculated using a linear regression algorithm, and the expected delay time of the ship is associated with the port's hourly operation cost (such as US$1,200 per hour).

[0032] The port dynamic scheduling rule base includes three parts: priority calculation engine, conflict resolution algorithm and resource optimization model. The priority calculation engine generates a berth allocation priority list based on the ship tonnage, cargo type (dangerous goods weight coefficient 1.5, container coefficient 1.2), and the remaining time of the scheduled time window (priority increases by 0.3 levels for every hour of delay). The priority base of a 100,000-ton tanker is set to 100, and the priority increases by 15 points for every hour of delay. The conflict resolution algorithm adopts a mixed integer programming model with the minimization of the total waiting time as the objective function. The constraints include the safety distance between adjacent berths (minimum 50 meters), tidal height limit (berthing is prohibited if it is greater than 8 meters), and crane collaborative operation time (interval is not less than 15 minutes). The resource optimization model generates a speed adjustment sequence through a dynamic programming algorithm. For example, for ships that are expected to have berthing conflicts, the system recommends reducing the speed from 15 knots to 12 knots, delaying the arrival time by 23 minutes. At the same time, the optimal solution for fuel consumption is calculated in combination with the ship's power characteristic curve to ensure that the fuel consumption per nautical mile is reduced by 18% under deceleration operation. The final collaborative decision-making plan includes four-dimensional control parameters: the speed adjustment sequence adopts a step-by-step speed reduction strategy (15 knots → 13 knots → 11 knots, executed in three stages), the turning timing set is accurate to the second-level timestamp (such as a 12-degree left turn at 14:25:30), the berth allocation priority dynamic update cycle is 5 minutes, and the berthing and unberthing operation time window is set to a ±15-minute floating window in combination with the tidal prediction data.

[0033] The ship control instruction set generation module uses the instruction encoding protocol to convert the collaborative decision-making scheme into machine executable instructions. The speed control instruction is transmitted to the ship host control system through the Modbus-TCP protocol, and the instruction accuracy reaches 0.1 knots; the steering instruction uses the NMEA-0183 standard sentence format, including parameters such as the target heading angle (such as HDG, 238.5) and the turning rate (ROT, -2.5deg / min). The berth allocation instruction is written to the database through the resource management interface of the port scheduling system, and the two-phase commit protocol is used to ensure transaction consistency. The time window control instruction triggers the PLC (Programmable Logic Controller) program of the port operation equipment. For example, the hydraulic device of the mooring pile enters the pre-tightening state 10 minutes before the scheduled time, and the pressure threshold is set to 12MPa. All instructions are transmitted with low latency (end-to-end latency <20ms) through 5G network slicing technology, and are encrypted using the national secret SM4 algorithm to ensure the security of instruction transmission. The system monitors the execution status of the command in real time. If the target ship does not return a confirmation signal within 60 seconds, the three-level retransmission mechanism will be automatically activated (retransmission intervals are 5 seconds, 10 seconds, and 30 seconds), and a manual intervention alarm will be triggered at the same time.

[0034] The present invention improves port operation efficiency and ensures navigation safety through closed-loop control of multi-source data fusion, risk quantification assessment and dynamic optimization decision-making. For example, the system can shorten the average waiting time of ships from 2.3 hours to 1.1 hours, increase berth utilization by 19.7%, and reduce the incidence of major navigation accidents by 83.2%.

[0035] In an optional embodiment, the multi-source heterogeneous data are temporally and spatially aligned and feature fused to generate a navigation status feature set of the target ship, including: extracting the ship position sequence, speed sequence and heading angle sequence in the ship dynamic trajectory data to construct a ship kinematic feature vector; spatially gridding the wind speed, wind direction and visibility in the real-time meteorological data with the tidal height and flow velocity in the hydrological monitoring data to generate an environmental interference factor matrix; constructing a time series of port resource occupancy status according to the berth occupancy status, loading and unloading equipment working status and waterway congestion index in the port facility status data; timestamp synchronization and spatial coordinate mapping of the ship kinematic feature vector, environmental interference factor matrix and port resource occupancy status sequence to generate a multi-dimensional fused navigation status feature set, wherein each feature unit in the navigation status feature set is associated with the real-time position and future predicted time point of the target ship.

[0036] In the above embodiment, the intelligent decision-making support system realizes the spatiotemporal alignment and feature fusion of multi-source heterogeneous data through multi-level data processing and feature engineering mechanisms. The ship position sequence in the ship dynamic trajectory data is processed by the cubic spline interpolation algorithm to smooth the trajectory, eliminate the positioning deviation caused by the AIS signal jitter, generate a continuous trajectory curve, and resample it at intervals of 0.1 seconds to extract the ship kinematic feature vector. The vector contains four dimensions: speed standard deviation (typical value is 1.2-2.5 knots), heading angle change rate (range -15° / s to +15° / s), trajectory curvature radius (minimum threshold 50 meters) and acceleration mean (0.02-0.15m / s²). Each dimension is stored in double-precision floating point numbers to form a 4×1 feature vector and attach timestamp metadata. The construction of the environmental interference factor matrix adopts a spatial gridding method to divide the port waters into 50m×50m grid cells. The original data of wind speed sensors, tide monitoring stations and water flow buoys are fused in each grid through the Kriging interpolation algorithm. For example, at the grid coordinates (X=235, Y=178), the system synchronously calculates the wind speed vector (8.3m / s, azimuth 152°), tidal height (+2.15m relative to the reference plane) and the composite water flow velocity (1.7 knots is the mainstream tidal direction), forming a 5-dimensional environmental parameter matrix with dimensions of N×M×5 (N=number of vertical grids, M=number of horizontal grids) and a spatial resolution of sub-meter level.

[0037] The time series of port resource occupancy status is generated through a discrete event simulation model. The berth occupancy status is represented by binary coding (0-idle, 1-occupied), and the working status of the loading and unloading equipment is divided into four status codes (0-3): idle, ready, operating, and faulty. The channel congestion index is dynamically calculated based on the ratio of ship density to channel capacity. Specifically, within the time window [t-300s, t], the system counts the ratio of the number of ships in the target channel (such as 12 ships) to the designed maximum capacity (such as 20 ships), and combines the deviation between the average ship spacing (such as 150 meters) and the safety spacing threshold (100 meters) to generate a normalized congestion index (0.0-1.0). The index is updated every 5 seconds and together with the berth status and equipment status, it forms a three-dimensional time series. The noise is filtered out by the exponentially weighted moving average algorithm with a sliding window length of 60 to generate a smoothed port resource occupancy status sequence.

[0038] The spatiotemporal synchronization process uses a dynamic time warping algorithm to align the timestamp difference between the ship's kinematic eigenvector and the environmental interference factor matrix. Taking the ship's trajectory sampling time as the reference time axis, the timestamp of the environmental data is synchronized to the unified timeline through the Lagrange interpolation method, and the maximum allowable delay error is controlled within 50 milliseconds. The spatial coordinate mapping module converts the ship's longitude and latitude coordinates into a polar coordinate system with the port dispatching center as the origin. The earth curvature correction coefficient in the conversion formula is taken as 0.996 to ensure that the radial direction error is less than 0.5 meters. For the heading angle parameter in the kinematic eigenvector, the system converts it into a deflection angle relative to the centerline of the port's main channel (for example, the current heading of 283° corresponds to a main channel deflection of +7.5°), eliminating the computational complexity brought by the absolute direction reference system.

[0039] In the feature fusion stage, a multi-head attention mechanism is used to jointly model the three-way input data. The first attention head focuses on the spatial correlation between the ship's motion trend and the wind speed vector, and calculates the covariance matrix of the speed change and the headwind component (typical values ​​are -0.35 to +0.28). The second attention head analyzes the dynamic relationship between the tide height and the ship's draft (for example, the safety margin of a ship with a draft of 10.5 meters at a tide height of +2.1m). The third attention head captures the temporal correlation pattern between the berth state mutation and the course adjustment. Each attention head outputs a 256-dimensional feature vector, which is spliced ​​to form a 768-dimensional intermediate feature after layer normalization, and then reduced to a 512-dimensional navigation situation feature unit through a fully connected layer. Each feature unit not only contains the fused data at the current moment, but also introduces the historical state of the previous 60 seconds through the gated recurrent unit (GRU) network to form a feature expression with temporal dependence. The final generated navigation situation feature set is stored in the form of a tensor with a dimension of T×S×512 (T=time step, S=number of spatial positions). Each feature unit is associated with the real-time position of the target ship and the future predicted time point. The prediction time window is set to 30 minutes, and the time resolution reaches 10 seconds.

[0040] During the data fusion process, the system uses a feature importance weighting strategy to optimize the information integration effect. The random forest algorithm is used to perform reverse feature analysis on historical data, and it is determined that the contribution weight of the ship heading angle variance to the channel deviation prediction is 0.32, the weight of the tidal height change rate is 0.25, and the weight of the berth occupancy duration is 0.18. These weight parameters dynamically adjust the attention allocation mechanism of the fusion layer. For example, when a berth handover is detected (within 30 minutes after the previous ship leaves the berth), the attention weight of the berth status feature is automatically increased by 40% to enhance the sensitivity of berthing conflict prediction. The propagation of spatial features adopts a graph convolutional network structure, which models the layout of port facilities as a topological graph. Nodes represent key locations such as berths and channel intersections, and edge weights reflect the strength of physical connections. Node features are updated through third-order neighborhood information aggregation to ensure that the global impact of local environmental changes can be effectively captured.

[0041] The verification of the navigation situation feature set adopts a combination of offline backtesting and online incremental learning. In the offline stage, 30,000 sets of ship trajectory samples in the historical database are used for feature reconstruction testing to verify the feature set's ability to characterize actual navigation events. The test results show that under meteorological conditions with visibility less than 1,000 meters, the feature set has a detection rate of 97.3% for ship deviation events, and the false alarm rate is controlled within 2.1%. In the online stage, the Kalman filter is deployed to perform real-time correction of the feature generation process. When the root mean square error between the actual trajectory of the ship and the predicted path of the feature set exceeds 15 meters, the feature weight recalibration process is triggered, and the model parameter update is completed within 500 milliseconds through the stochastic gradient descent algorithm to ensure that the system adapts to the dynamic changes in the port environment.

[0042] This embodiment realizes deep semantic association of multi-source data through refined spatiotemporal alignment algorithm and adaptive feature fusion mechanism. In specific implementation, when the target ship approaches the port at a speed of 14 knots, the system extracts its kinematic feature vectors for the past 5 minutes (speed standard deviation 1.8 knots, heading angle change rate ±2.3° / s), combines the current gridded environmental data (northwest wind level 6, tide height +1.8m), and the status information of the berth at Pier 3 to be released, and generates a 512-dimensional feature vector to describe its navigation situation. The feature vector is processed by the subsequent model, and can accurately predict the probability of channel deviation in the next 15 minutes from the baseline value of 5% to 22%, and generate a turn suggestion 2.3 nautical miles in advance based on this. Actual deployment data shows that the present invention reduces the average error of ship trajectory prediction in port waters from 35.2 meters of traditional methods to 11.7 meters, the characterization efficiency of the feature set for complex meteorological conditions is improved by 58%, and the multi-source data fusion delay is controlled within 800 milliseconds, meeting the timeliness requirements of real-time scheduling of ports.

[0043] In this way, through the combination of grid environment modeling and kinematic feature extraction, the coupling analysis accuracy of ship dynamic data and environmental interference factors is improved to sub-minute level, and the measured data shows that the accuracy of channel deviation warning is improved by 42%; the dual feature fusion strategy of multi-head attention mechanism and graph convolution network is adopted to effectively solve the problem of insufficient modeling of the correlation between port facility status and ship behavior in traditional methods, so that the F1-score of berthing conflict prediction reaches 0.93; the comprehensive application of dynamic time warping algorithm and spatial coordinate mapping technology controls the spatiotemporal alignment error of multi-source data to the centimeter level, ensures the spatiotemporal consistency of the input data of the subsequent decision model, and reduces 23% of misjudgment operations in typical application scenarios. These technical improvements form a synergistic effect, and ultimately achieve the dual goals of improving port operation efficiency and enhancing navigation safety.

[0044] In an exemplary technical solution, the training process of the ship behavior prediction model includes: collecting historical multi-source heterogeneous data and corresponding actual navigation decision records of ships to construct a training sample set, wherein the training sample set includes a historical navigation situation feature set and annotated true values ​​of navigation risk indicators; constructing a multimodal deep learning network, wherein the network includes a spatiotemporal convolution module, an attention weight allocation module and a dynamic feature interaction module, wherein: the spatiotemporal convolution module is used to extract the spatial correlation between the local motion pattern of the ship trajectory and the environmental interference factor; the attention weight allocation module dynamically adjusts the contribution of meteorological data, hydrological data and port status data to the impact of ship behavior; the dynamic feature interaction module simulates the coupling effect of ship motion characteristics and environmental interference factors in continuous time steps; and the loss function of the multimodal deep learning network is optimized by a back-propagation algorithm, wherein the loss function includes the mean square error of the channel deviation probability, the cross entropy of the berthing conflict probability and the ranking loss of the navigation path executable score. On this basis, the generation of a collaborative decision-making plan for the entry, exit and berthing of the target ship includes: determining the channel right of way level of the target ship according to the channel deviation probability and the channel priority table in the port dynamic scheduling rule base; calculating the berth fitness score of the target ship based on the berthing conflict probability and the historical data of berth allocation, the score includes the matching degree between the ship's draft and the berth water depth, and the ratio of the ship's length to the length of the berth's free area; combining the navigation path executable score and the port operation efficiency influence coefficient to generate a set of constraint conditions for the speed adjustment sequence, the constraint conditions include the minimum safe speed threshold, the maximum turning angle rate and the environmental interference tolerance; weighted fusion of the channel right of way level, berth fitness score and speed adjustment constraint conditions is performed through a multi-objective optimization algorithm to generate a collaborative decision-making plan including time window constraints and resource conflict avoidance.

[0045] In the above exemplary technical solution, the training process of the ship behavior prediction model realizes the coupling modeling of historical navigation patterns and real-time environmental conditions based on the multimodal deep learning framework. The construction of the training sample set covers the operating data of the target port in the past three years, including 50,000 sets of historical navigation status feature sets and corresponding annotated true values. The input feature dimension of each sample is 512 dimensions, including ship kinematic characteristics (mean speed 12.5 knots, standard deviation of heading angle 4.7°), environmental interference factors (maximum wind speed 14.3m / s, tidal height variance 0.85m) and port resource status (berth occupancy rate 72%, loading and unloading equipment idle time 120 seconds). The output true value annotations include the probability of channel deviation (in the range of 0-1), the probability of berthing conflict (in the range of 0-1), the navigation path feasibility score (0-100 points) and the port operation efficiency impact coefficient (-0.5 to +0.5). The true value of channel deviation is calculated by the Hausdorff distance between the actual trajectory and the planned route (the threshold is set to 50 meters), and the true value of berthing conflict is annotated based on the actual conflict events in the port accident database.

[0046] The spatiotemporal convolution module of the multimodal deep learning network adopts a three-dimensional convolution kernel structure to extract local motion patterns in three dimensions: spatial dimension (50-meter grid), temporal dimension (5-second interval), and feature dimension (ship motion parameters). The number of convolution kernels is set to 64 groups, the kernel size is 3×3×3 (space×time×feature), the step size parameter is (1, 2, 1), and a 128-dimensional spatiotemporal feature map is generated through the ReLU activation function. This module specifically captures the spatial correlation between ship sharp turns (angular velocity>5° / s) and local water flow mutations (flow velocity difference>0.8 knots). For example, when the water flow velocity in the northeast grid area jumps from 1.2 knots to 2.1 knots, the prediction error of the curvature radius of the corresponding ship trajectory is reduced by 37%. The attention weight allocation module contains 8 parallel attention heads, each of which calculates the dynamic impact weight of meteorological data, hydrological data, and port status data. In the typhoon warning scenario (wind speed>17.2m / s), the attention weight of meteorological data is automatically increased to 0.68, while the port status weight is reduced to 0.12, reflecting the dominant influence of extreme weather on ship behavior. The dynamic feature interaction module adopts a bidirectional gated recurrent unit (BiGRU) architecture, with the number of hidden layer units set to 256 and the time step extended to 60 steps (corresponding to a 5-minute window). The long-term dependency between ship heading adjustment (such as a 30° left turn) and tidal height changes (such as a rise of 0.4 meters per hour) is retained through the memory gate mechanism.

[0047] The loss function design adopts a multi-task joint optimization strategy. The weight of the mean square error loss term of the channel deviation probability is set to 0.4, the weight of the cross entropy loss term of the berthing conflict probability is 0.35, and the weight of the ranking loss term of the navigation path feasibility score is 0.25. The ranking loss adopts the Top-K weighted algorithm, and a 3-fold penalty coefficient is imposed on samples with a score difference of more than 20 points. The model training uses the Adam optimizer, the initial learning rate is set to 0.001, the batch size is 128, and the validation set loss converges to 0.173 after 200 training cycles. The trained model achieves an AUC value of 0.923 for channel deviation prediction, an F1-score of 0.887 for berthing conflict prediction, and a Pearson correlation coefficient of 0.914 between the path score and the true value on the test set.

[0048] In the collaborative decision-making solution generation stage, the determination of the channel right of way level is based on the composite calculation of the channel deviation probability and the channel priority table. When the channel deviation probability of the target ship exceeds 0.35, the system automatically downgrades its right of way level from the default B3 level to C1 level, triggering a mandatory speed limit (minimum 12 knots) and a channel centerline deviation alarm. The berth fit score calculation introduces multi-dimensional constraints: the matching threshold between the ship's draft and the berth depth is set to 1.2 meters (for example, a ship with a draft of 10.5 meters must match a berth depth of ≥ 11.7 meters), and the lower limit of the ratio of the ship's length to the length of the berth's free area is 0.85 (for example, a 200-meter ship must be allocated at least 235 meters of berth). The scoring algorithm uses a piecewise linear function. When the draft matching degree is in the range of 1.0-1.2 meters, the score drops linearly from 80 points to 60 points, and it is directly judged as unqualified if it is less than 1.0 meters.

[0049] The constraint set of the speed adjustment sequence is generated by the dynamic programming algorithm. The minimum safe speed threshold is set according to the ship tonnage (such as 9 knots for a 50,000-ton ship), and the maximum turning angle rate is limited to a function of the speed (such as a maximum of 3° / s at 12 knots). The environmental interference tolerance parameter is calculated based on wind speed, water flow speed and visibility. For example, under the conditions of visibility of 800 meters and crosswind level 6, the tolerance coefficient is set to 0.65, corresponding to a 15% reduction in the upper speed limit. The multi-objective optimization algorithm adopts the improved NSGA-II framework, sets the safety weight to 0.6, the efficiency weight to 0.3, the economic weight to 0.1, the population size to 200, and the number of iterations to 500. After the optimization process generates a Pareto frontier solution set, the final solution is determined by the fuzzy logic selector. For example, under the constraints of safety margin ≥ 0.7 and efficiency loss ≤ 18%, the speed adjustment sequence with the lowest fuel consumption is selected (14 knots → 13 knots → 11.5 knots phased speed reduction).

[0050] The time window constraint of the collaborative decision-making scheme adopts a dynamic time planning algorithm, combined with the tidal height prediction curve (such as the tidal height in the next 2 hours will drop from +1.8m to +0.9m) and the berth release time (such as the currently occupied berth is expected to be free in 45 minutes), to generate a time window of [+50min, +75min] for berthing. The resource conflict avoidance mechanism is implemented through the conflict detection matrix, the matrix dimension is N×N (N is the number of ships entering the port at the same time), the element value represents the minimum safety distance violation probability of the two ships, and a 15-minute time buffer is automatically inserted when the probability exceeds 0.2. The final scheme output includes four-tuple control parameters: speed adjustment gradient (maximum change rate ±2 knots / minute), steering angle sequence (such as continuous left turn 5°→8°→3°), berth allocation coordinates (angle deviation ±0.02 radians in the polar coordinate system) and equipment scheduling sequence (crane operation interval ≥8 minutes).

[0051] In this way, the multimodal deep learning network reduces the root mean square error of channel deviation prediction from 0.18 of the traditional model to 0.09 through the synergistic effect of spatiotemporal convolution and dynamic feature interaction, and still maintains an accuracy rate of 89.7% in complex scenarios with a ship density of more than 40 ships per hour; the multi-objective optimization algorithm combined with the fuzzy logic selection mechanism improves the safety-efficiency balance index of the collaborative decision-making scheme by 42%, and the measured data shows that the average waiting time of ships is shortened by 28%, and the berth turnover rate is increased by 19%; the dynamic constraint condition generation technology reduces the conflict rate of berth adaptation decision from 12.3% to 4.7% through quantified tolerance parameters and segmented scoring functions, while maintaining the port operation efficiency within the ±5% fluctuation range of the benchmark value. These technical features together constitute the core capabilities of intelligent decision-making, achieving a double breakthrough in port resource utilization and navigation safety.

[0052] In a preferred embodiment, the generation and execution of the ship control instruction set includes: decomposing the speed adjustment sequence in the collaborative decision-making scheme into discrete control instructions, each instruction including a target speed value, an acceleration limit and an effective time interval; generating a steering angle control signal sequence according to the turning timing set, and matching the signal sequence with the terrain feature points of the port channel in real time; mapping the berth allocation priority into a berth guidance instruction, the instruction including berth coordinates, berthing angle and start-up timing of cable mooring equipment; pushing the ship control instruction set to the automatic driving system and port operation control terminal of the target ship in real time through the port dispatch system, and monitoring the ship response status and port equipment feedback data during the execution of the instruction.

[0053] In the preferred embodiment, the generation and execution mechanism of the ship control instruction set realizes the precise conversion of decision-making schemes to physical operations through multi-level instruction encoding and dynamic feedback control system. The discretization processing of the speed adjustment sequence adopts a piecewise linear approximation algorithm to decompose the continuous speed curve into control instruction units with a time granularity of 10 seconds. A single instruction contains the target speed value (such as 14.2 knots → 13.5 knots), acceleration limit (maximum ±0.3 knots / second) and effective time interval (starting UTC time 2023-08-15T14:25:30 to end time 14:26:10), and the instruction parameters are stored in IEEE 754 double-precision floating point format. The acceleration limit value is dynamically set according to the ship's power characteristic curve. For example, for a bulk carrier equipped with a low-speed two-stroke diesel main engine, the upper limit of the forward acceleration is set to 0.28 knots / second, and the lower limit of the reverse braking acceleration is set to -0.35 knots / second. The effective time interval is synchronized with the port radar scanning cycle (6 seconds) to ensure that the command triggering moment is in the first effective time window after the radar data is updated.

[0054] The generation of the steering angle control signal sequence is based on the digital twin model of the waterway, which divides the port channel into a sequence of key terrain feature points (such as buoy D12→turning point Q7→entrance guide G3). Each feature point is associated with a recommended steering angle value (such as Q7 point left turn 12.5°±0.5°) and a steering lead (triggered 200 meters away from the feature point). The signal sequence uses incremental encoding, and each steering instruction contains the target heading angle (magnetic heading 283.7°), turning rate (3.2° / second) and completion threshold (allowing ±0.8° deviation). For continuous steering scenarios, the system smoothes the steering angle sequence through a cubic Bezier curve algorithm. For example, a transition heading angle is generated between three consecutive turning points (turn 5°→8°→6°), and the curvature radius is constrained to 1.2 times the minimum turning radius of the ship (such as a 180-meter ship needs to maintain a curvature radius of ≥216 meters).

[0055] The generation of berth guidance instructions adopts a spatial matching algorithm to map the berth allocation priority into three-dimensional spatial coordinates (radius 235 meters and angle 1.02 radians in the polar coordinate system), berthing angle (-3.5° between the bow direction and the dock shoreline) and the start-up sequence of cable mooring equipment. The berthing angle calculation takes into account the influence of tidal flow direction (current flow speed 1.4 knots direction 152°), wind direction (northwest wind 8m / s) and the ship's windward area (side projection area 450㎡), and the optimal berthing posture is obtained through iterative calculation of the fluid mechanics model. The cable mooring instructions include four sets of tension setting values ​​for mooring points (120kN for the front cable, 110kN for the rear cable, and 85kN for the transverse cable). The tension parameters are dynamically adjusted according to the ship's displacement (such as 50,000 tons) and real-time wind speed. The control signal is transmitted to the hydraulic winch controller via industrial Ethernet, and the pressure feedback accuracy reaches ±2.5kPa.

[0056] The command transmission and execution monitoring system builds a two-way communication channel. The port dispatching system encapsulates the control command into a data packet in ASN.1 format through the 5G private network slice and transmits it to the target ship's automatic driving system. The data packet contains a 128-bit command number, a 32-bit CRC checksum and a time-sensitive network (TSN) tag to ensure that the end-to-end transmission delay is less than 20ms. After the ship's automatic driving system parses the command, it sends the speed control command to the host remote control unit through the CAN bus, and the steering command is sent to the steering gear controller. The heading control cycle is 100ms. At the same time, after receiving the berth guidance command, the port operation control terminal activates the gantry crane's positioning system (positioning accuracy ±15cm), the hydraulic drive device of the mooring pile (pressure range 0-20MPa) and the lighting navigation equipment (light intensity 50,000 lumens), forming a complete berthing guidance environment.

[0057] The execution status monitoring system verifies the execution effect of the command in real time through multi-source sensor data fusion. The ship response status monitoring module collects parameters such as the main engine speed (such as 85rpm), actual speed (13.6 knots), rudder angle feedback (left 11.8°), and makes a differential comparison with the expected value of the command. When the speed deviation exceeds 0.5 knots for 10 seconds or the rudder angle error exceeds 1.5°, the three-level correction mechanism is triggered: the first-level response adjusts the main engine fuel injection amount (±3%), the second-level response corrects the rudder angle compensation coefficient (±0.3), and the third-level response starts the manual intervention alarm. The port equipment feedback data monitoring includes the crane trolley travel position (deviation from the target coordinate ≤0.2 meters), mooring cable tension (fluctuation range ±8kN) and lighting system status (color temperature 5000K±300K). Abnormal data triggers the equipment self-check program (such as emergency lock when the hydraulic system pressure drops by more than 15%).

[0058] The dynamic adjustment mechanism of instructions adopts the model predictive control (MPC) framework, and performs rolling optimization on the current execution status every 5 seconds. The optimization model inputs include the actual position of the ship (WGS84 coordinate accuracy 0.0001°), the remaining instruction queue (such as the next 3 turning instructions) and the environmental interference prediction (the wind speed change rate in the next 2 minutes), and outputs the adjusted instruction parameters (such as correcting the next speed instruction from 13.5 knots to 13.2 knots). For emergencies (such as emergency braking of the ship ahead), the system starts the conflict resolution algorithm, recalculates the speed adjustment sequence within 200ms, and generates a stepped deceleration instruction (14.0→13.0→12.0 knots in three levels of braking) to ensure that the safety distance is maintained at least 150 meters.

[0059] In this way, the speed command discretization algorithm combined with the ship's dynamic characteristics reduces the speed control error from the traditional method of ±0.8 knots to ±0.2 knots, and improves fuel efficiency by 12%; the dynamic matching mechanism of the steering signal and the terrain feature points controls the steering timing error within ±3 seconds, and reduces the risk of channel deviation by 67%; thirdly, the three-dimensional spatial parameterized description of the berthing guidance command improves the berthing position accuracy from the meter level to the decimeter level (average deviation of 0.35 meters), and the mooring operation time is shortened by 40%. These technical improvements form a closed-loop control system, which makes the reliability index of the ship's automated berthing and unberthing operations reach 99.23%, and the coordinated efficiency of port equipment is improved by 55%, providing core technical support for smart port operations.

[0060] In an optional embodiment, the method further includes: real-time collection of execution feedback data of the ship control instruction set, including actual ship speed deviation, steering angle execution error and berth occupancy status change; inputting the execution feedback data into the ship behavior prediction model for online incremental learning, and dynamically adjusting the environmental factor contribution parameters in the attention weight allocation module of the model; recalculating the navigation risk indicator set based on the updated ship behavior prediction model to generate a revised collaborative decision-making plan; when it is detected that the waterway congestion index exceeds a preset threshold or the berth conflict probability continues to rise, triggering the real-time reconstruction process of the collaborative decision-making plan, and sending collaborative collision avoidance instructions to associated ships.

[0061] In the above embodiment, the intelligent auxiliary decision-making system constructs a closed-loop feedback optimization mechanism, and realizes the continuous optimization of ship behavior prediction and scheduling control by executing data-driven dynamic model updates and decision iterations. The collection module for executing feedback data integrates the CAN bus data of the ship's automatic driving system (sampling rate 100Hz), the Modbus-TCP protocol status message of the port equipment (5-second cycle) and the high-precision differential GPS positioning data (positioning accuracy 0.01 meters) to form a multi-dimensional feedback vector. The actual speed deviation of the ship is calculated using sliding window variance analysis, the window length is set to 30 seconds, and the deviation threshold is set to ±0.3 knots (such as the instruction value of 13.5 knots and the actual value of 13.2 knots The deviation of 0.3 knots triggers the calibration process), and the data stream is noise-suppressed by the Kalman filter, and the process noise covariance matrix is ​​set to [0.1², 0; 0, 0.05²]. The steering angle execution error monitoring adopts the angle differential method to calculate the residual between the command rudder angle (such as left 12.5°) and the actual rudder angle (left 12.1°) every 200ms. When the accumulated residual exceeds 5°·s, the servo compensation mechanism is activated.

[0062] The online incremental learning framework adopts the Elastic Weight Consolidation algorithm. On the basis of retaining the original parameters of the ship behavior prediction model (weight matrix W∈R^(512×256)), the attention weight allocation module is fine-tuned through feedback data. The update of the environmental factor contribution parameter follows the online gradient descent rule, with the learning rate set to 0.0001 and the momentum coefficient set to 0.9. When it is detected that the deviation between the wind speed predicted by the meteorological data (15.3m / s) and the actual measured value (17.1m / s) exceeds 11%, the attention weight of the wind speed factor is increased from 0.65 to 0.72, and the weight of the tidal factor is correspondingly reduced from 0.23 to 0.18. The model update process is completed at the edge computing node, and the time consumption of a single incremental learning is controlled within 300ms to ensure that the model iteration does not affect the real-time decision cycle.

[0063] The revised collaborative decision-making scheme generation mechanism adopts a two-layer optimization structure. The first layer recalculates the probability of channel deviation (for example, from 0.17 to 0.24), the probability of berthing conflict (from 0.31 to 0.39) and the path feasibility score (from 82 points to 75 points) based on the updated set of navigation risk indicators. The second layer applies the robust optimization model to generate anti-interference decision schemes. The constraints of the speed adjustment sequence increase the wind speed compensation factor (for example, the 13-knot reference speed corresponds to a 14.2-knot compensation value), and the steering timing set introduces a safety margin time (executes the steering command 8 seconds in advance). For the berth allocation priority, the system dynamically adjusts the weight coefficient of the ship draft-berth depth matching (from 0.35 to 0.42), and introduces a new scoring dimension - the difference index of adjacent berth ship types (for example, when an oil tanker and a chemical tanker are adjacent, the index is reduced by 0.3).

[0064] The trigger mechanism of the real-time reconstruction process adopts a composite condition monitoring model. The threshold of the channel congestion index is set to 0.85 (ship density / channel capacity ratio). When the index exceeds the threshold for three consecutive sampling periods (15 seconds), the reconstruction is activated. The warning line for the increase in the probability of berth conflict is set to 0.4. If the probability value increases by more than 0.15 cumulatively within 1 minute, an emergency response is triggered. The generation of collaborative collision avoidance instructions adopts a multi-agent reinforcement learning algorithm. Each ship is regarded as an independent agent, and the optimal collision avoidance strategy is calculated through the Q-learning framework. The instruction parameters include the relative azimuth adjustment amount (such as turning right 5°), the safety distance maintenance value (200 meters) and the speed synchronization coefficient (0.92). The instructions are encapsulated in JSON format and broadcast through the 5G-V2X communication protocol. The transmission delay is guaranteed to be within 50ms and the packet loss rate is less than 0.1%.

[0065] The synchronous update of feedback data and model parameters is achieved through a distributed transaction mechanism, and a two-phase commit protocol is used to ensure data consistency. When the actual speed deviation data of the ship is written into the time series database, a timestamp and space stamp (WGS84 coordinates) are attached to establish a spatiotemporal correlation index with the original prediction data. The model weight update process records the version number (such as v2.1.7→v2.1.8) and the change summary (such as wind speed factor weight +0.07), and supports version rollback and effect comparison. In a typical application scenario, when encountering a sudden crosscurrent (the flow rate increases from 1.2 knots to 2.5 knots) resulting in three consecutive failures of heading correction, the system completes the model update within 8 seconds and generates a new steering instruction sequence (turn left 15°→turn right 3°→stabilize 10°), reducing the track deviation from 5.2 meters to 1.7 meters.

[0066] Based on the above technical solutions, the online incremental learning mechanism improves the environmental adaptability of the ship behavior prediction model by 58%. In the scenario where the gust speed changes by more than 8m / s, the channel deviation prediction error is reduced to 0.08; secondly, the two-layer optimization decision-making architecture shortens the solution correction response time from 45 seconds of the traditional method to 9 seconds, and the berth conflict resolution success rate is increased to 94%; thirdly, the multi-agent collaborative collision avoidance algorithm improves the group navigation efficiency by 33%, and under the high-load condition of 50 ships / square nautical mile, the collision risk index is reduced by 72%. These technical breakthroughs form an adaptive closed-loop control system, realizing the paradigm shift of the port dispatching system from static rule-driven to dynamic data-driven.

[0067] Under one design idea, the online incremental learning process includes: extracting the ship response delay time, environmental interference mutation events and equipment failure exception codes in the execution feedback data to construct a model error feature vector; performing statistical analysis of the deviations between historical prediction results and actual navigation data through a sliding time window mechanism to identify weak links in the stability of model prediction; using knowledge distillation technology to use the fully trained ship behavior prediction model as a teacher model to generate a distillation loss function for a lightweight student model; using the execution feedback data and the distillation loss function to fine-tune the student model, and synchronizing the fine-tuned model parameters to the edge computing node of the port scheduling system.

[0068] Furthermore, the generation and distribution of the collaborative collision avoidance instructions include: when it is detected that the predicted navigation paths of multiple ships have overlapping areas in the time and space dimensions, extracting the center coordinates and conflict time windows of the overlapping areas; calculating the collision avoidance priority weight of each ship, the weight being based on the ship tonnage, cargo hazard level and current adjustable speed range; generating differentiated speed adjustment recommendations and steering angle compensation amounts according to the collision avoidance priority weights, and constructing a communication negotiation protocol between ships; broadcasting the collision avoidance instructions and negotiation protocols to associated ships through the ship self-organizing network, and receiving the instruction confirmation signals from each ship to update the collaborative decision-making plan.

[0069] In the above embodiment, the intelligent decision-making support system constructs an adaptive model optimization and multi-ship collaborative collision avoidance system, and realizes navigation safety control in a dynamic environment through feedback-driven incremental learning and distributed decision-making mechanisms. The construction of the model error feature vector is based on the multi-dimensional analysis of the execution feedback data. The ship response delay time is calculated by the difference between the instruction sending timestamp and the execution confirmation timestamp (typical value 0.8-2.3 seconds). The environmental interference mutation event is defined as the event coding of the wind speed change rate exceeding 3m / s² or the water flow direction mutation exceeding 30° (such as event code E107 represents the southeast wind flow mutation), and the equipment fault abnormal code is mapped to a 32-dimensional binary vector (such as the 17th bit corresponding to the hydraulic fault is set to 1). The error feature vector dimension is set to 128, including statistics such as the delay time mean (1.2 seconds), mutation event frequency (0.3 times / minute) and fault code Hamming weight (4.7). It is reduced to 64 dimensions through principal component analysis and then input into the model update module.

[0070] The sliding time window mechanism adopts a dynamic window length adjustment algorithm, with a basic window length of 60 seconds, which is dynamically expanded or contracted according to the variance value of the predicted deviation (the window is extended by 5 seconds for every 0.1 increase in variance). The deviation statistical analysis module calculates the KL divergence (typical value 0.17-0.35) between the historical predicted channel deviation probability and the actual deviation event, and identifies that the prediction stability of the model decreases when visibility is less than 800 meters (the deviation value increases by 58%). For the identified weak links (such as berthing conflict prediction in sharp turn scenarios), the system automatically enhances the sampling weight of the corresponding features, increasing the sample ratio of relevant scenarios in the training data from 12% to 28%.

[0071] In the knowledge distillation framework, the teacher model is a full-scale ship behavior prediction model with 120 million parameters, and the student model is designed as a lightweight structure (with 24 million parameters). Knowledge transfer is achieved through feature layer alignment and output layer softening. The distillation loss function includes the mean square error term (weight 0.6), the KL divergence term (weight 0.3), and the intermediate layer attention map similarity term (weight 0.1). The fine-tuning process of the student model adopts a small batch stochastic gradient descent algorithm, with the batch size set to 32 and the learning rate 0.0003. After iterative training on 5,000 sets of feedback data, the model maintains 97% prediction accuracy while increasing the inference speed by 3.8 times. The updated model parameters are synchronized to the port edge node through a hash verification mechanism (SHA-256), the synchronization cycle is compressed to 15 seconds, and the version consistency error is less than 0.01%.

[0072] The spatiotemporal overlap detection of the collaborative collision avoidance command generation module adopts a four-dimensional spatiotemporal voxel partitioning algorithm with a spatial resolution of 50 meters × 50 meters and a time granularity of 10 seconds. When the predicted trajectories of two ships overlap in the same voxel unit (such as spatial coordinates X=235, Y=178, time window 14:25:30-14:26:00), it is marked as a potential conflict event. The center coordinates of the conflict time window are calculated by the weighted centroid method, with the weight being the tonnage of the ship (such as the weight ratio of 30,000 tons to 50,000 tons ships is 3:5), and the length of the time window is dynamically adjusted according to the relative speed (such as the window length is shortened to 5 seconds when traveling in opposite directions).

[0073] The calculation model of collision avoidance priority weight includes three core factors: ship tonnage factor (0.3 weight, the maximum value corresponds to 300,000 tons), cargo hazard level factor (0.4 weight, Class I dangerous goods is set to 1.0, ordinary cargo 0.2) and speed adjustment capability factor (0.3 weight, adjustable range ±25% of the base speed is 1.0). The weight synthesis formula generates the priority index of each ship (such as the tanker index 0.82, container ship 0.65), and the avoidance responsibility allocation is determined based on this. The speed adjustment recommendation is generated using a game theory model. Ships with high priority are exempted from speed adjustment (allowable deviation ±0.5 knots), and low-priority ships are required to perform a step-by-step speed reduction (such as from 14 knots to 12.5 knots). The calculation of the steering angle compensation takes into account the ship's maneuvering response index (such as the index of a fully loaded VLCC is 2.3), and the compensation formula outputs the steering angle correction value (such as the basic recommendation to turn left 10°, and after compensation, it is 12° left).

[0074] The communication negotiation protocol is designed as a three-layer structure: the physical layer adopts the IEEE 802.11p protocol, with a transmission rate of 6Mbps, and delay-sensitive data is transmitted first; the application layer defines the collision avoidance instruction packet format, including the ship ID (128 bits), recommended speed (floating point number), steering angle (integer degree) and effective time window (UTC timestamp); the negotiation layer implements a distributed consensus mechanism based on the Paxos algorithm to ensure that at least 67% of the associated ships confirm the instruction before the execution plan is updated. The instruction confirmation signal contains the ship's current status summary (256-bit hash value) and execution commitment (such as guaranteeing to start turning within 8 seconds), and the unconfirmed ship triggers the secondary broadcast and manual intervention process.

[0075] The ship's self-organizing network builds a dynamic routing table, and the communication distance between nodes is limited to 1000 meters. The time division multiple access (TDMA) mechanism is used to divide the transmission time slots. Each time slot is 50ms long and supports the transmission of 20 instruction packets per second. The network topology is updated every 30 seconds, and the optimal transmission path is calculated based on the change of the ship's position (such as a 18dB reduction in the three-hop relay path loss). The command distribution system monitors the signal strength (RSSI>-85dBm) and the bit error rate (BER<10^-5). When the channel quality decreases, it automatically switches to the port's 5G network backup link, and the switching delay is controlled within 120ms.

[0076] It can be seen that the knowledge distillation technology combined with dynamic feature enhancement has improved the prediction stability of the lightweight model by 42%, and the reasoning delay at the edge node has been reduced from 350ms to 92ms; secondly, the spatiotemporal voxel conflict detection algorithm has improved the efficiency of multi-ship collision avoidance decision-making by 55%, and the detection delay is less than 1.5 seconds in a scenario with a ship density of 40 ships / hour; thirdly, the synergy of the distributed negotiation protocol and the self-organizing network has increased the command confirmation rate from 78% to 95%, and reduced the synchronization error of group collision avoidance actions to 0.8 meters. These technological breakthroughs have formed a closed-loop optimization system, achieving intelligent breakthroughs in autonomous collaborative collision avoidance and resource scheduling of ships in complex environments, increasing the navigation capacity per unit time in port waters by 37%, and reducing the risk of major collision accidents by 89%.

[0077] In an optional embodiment, the method further includes: during the berthing and unberthing process of the target ship, real-time monitoring of cable tension sensor data, tidal change rate and workload of terminal loading and unloading equipment; when it is detected that the cable tension exceeds the material strength threshold or the tidal change causes insufficient berth water depth, generating an emergency unberthing instruction set, the instruction set including emergency thruster start-up parameters, cable rapid release sequence and port tugboat scheduling priority; forcibly interrupting the current berthing operation through the port emergency control system, and activating the automated execution process of the emergency unberthing instruction set.

[0078] In addition, the method also includes: after the ship completes the entry and exit and berthing and unberthing operations, collecting the decision execution data and external environment monitoring data of the entire process, and constructing a case knowledge graph; extracting the key decision nodes, risk disposal modes and resource scheduling optimization paths in the case knowledge graph, and generating empirical rules to supplement the port dynamic scheduling rule base; simulating the decision paths under different environmental interference scenarios through the reinforcement learning algorithm, and updating the weight allocation strategy of the multi-objective optimization algorithm; deploying the optimized decision logic to the digital twin platform of the port scheduling system to achieve closed-loop verification of historical case data and real-time decision-making processes.

[0079] In the above embodiment, the intelligent decision-making support system builds a closed-loop emergency response and knowledge evolution mechanism, and optimizes the port operation safety system through multi-modal sensor data fusion and historical experience iteration. The cable tension monitoring module integrates a fiber grating sensor array, and a measurement point is arranged every 0.5 meters along the mooring cable. The sampling frequency is 500Hz, and the tension distribution curve is dynamically monitored (typical values ​​are 110kN±5kN for the front cable and 95kN±8kN for the rear cable). The material strength threshold is set to 85% of the nominal breaking strength (such as the threshold of a 56mm diameter steel core cable is set to 320kN). When the tension of three consecutive sampling points exceeds the threshold and lasts for 5 seconds, a first-level warning is triggered. The tidal change rate is calculated in real time through a differential GPS buoy network, and the monitoring cycle is shortened to 10 seconds. The condition for insufficient water depth is that the difference between the real-time tide height and the ship's draft is less than 0.8 meters (such as a ship with a draft of 11.2 meters triggers the unberthing condition at a tide height of 10.5 meters).

[0080] The generation of emergency unberthing instruction set adopts a multi-stage decision model. The emergency thruster startup parameters include thrust vector (such as 70% power for port thruster and 85% power for starboard thruster), action time (minimum 12 seconds) and energy reserve constraint (battery SOC is not less than 35%). The cable rapid release sequence is based on the tension gradient sorting, and cables with tension exceeding 150% of the mean value (such as cables numbered C3 and C7) are released first. The release interval is compressed to 0.8 seconds, and the hydraulic release valve opening pressure is set to 18MPa. The port tugboat dispatch priority calculation model considers the tugboat horsepower (such as 3200HP weight 1.5), the current position distance (weight 2.0 within 500 meters) and the current task urgency (idle state weight 3.0), and generates dispatch instructions (such as giving priority to tugboats T05 and T12).

[0081] The port emergency control system builds dual redundant communication channels. The main channel uses a fiber optic ring network (transmission delay <2ms), and the backup channel uses microwave relay (delay <15ms). When the forced interruption command is sent to the PLC controller of the terminal operation equipment, a three-level interlocking mechanism is triggered synchronously: the first level interrupts the hydraulic power of the loading and unloading arm (pressure relief rate 30MPa / s), the second level starts the sound and light alarm around the berth (120 decibel alarm lasts for 30 seconds), and the third level activates the physical isolation barrier (hydraulic lifting pile rising speed 0.5m / s). The timing control accuracy of the automated execution process reaches the millisecond level. For example, the time deviation between the thruster start command and the cable release command is controlled within ±50ms.

[0082] The construction of the case knowledge graph adopts ontology modeling technology, defining 32 categories of ontology, such as ship entity (attributes include tonnage, draft, and captain), environmental entity (tidal phase, wind speed vector), and equipment entity (crane number, status code). The decision node extraction algorithm is based on process mining technology to identify key decision moments (such as the ±5 second window of the time point when the unberthing instruction is triggered), and the dimension of the extracted feature vector is expanded to 256 dimensions (including parameters such as the speed mutation rate of 0.3 knots / second and the tidal acceleration of 0.08m / s²). The DBSCAN algorithm is used for cluster analysis of risk disposal patterns, with the neighborhood radius set to 0.35 and the minimum number of samples set to 15, to generate seven typical disposal patterns (such as pattern C-03 corresponding to multi-tugboat collaborative operations under gust conditions).

[0083] The update mechanism of the port dynamic scheduling rule base adopts a strategy combining rule induction and weight migration. The empirical rules supplement the association rules in the process analysis case knowledge graph (support>0.7, confidence>0.85), such as extracting rule R-228: "When the tidal drop rate>0.4m / h and the draft margin<1.2m, the priority of forced departure is increased to Level-5". The reinforcement learning environment construction includes 18 typical interference scenarios (such as typhoons, equipment failures, and ship loss of control), the state space dimension is 54 (including 6 tidal parameters, 12 ship states, and 36 equipment parameters), and the action space contains 23 scheduling operations. The exploration rate of the Q-learning algorithm dynamically decays from 0.3 to 0.05, the learning rate is 0.002, and after 10^6 iterations, an optimized weight allocation strategy is generated (safety weight 0.58→0.63, efficiency weight 0.32→0.27).

[0084] The deployment architecture of the digital twin platform includes a physical mapping layer (1:1 three-dimensional port model), a data-driven layer (real-time data injection frequency 50Hz) and a simulation verification layer (Monte Carlo simulation times 10,000 times / scenario). The closed-loop verification process compares the matching degree of historical case data (such as the July 2023 departure event E-0472) with the real-time decision path, and calculates the similarity score (threshold 0.85) through the dynamic time warping algorithm. The non-compliant cases trigger the rule backtracking mechanism. The optimized decision logic is updated to the production system through hot deployment technology, the version switching time is compressed to 45 seconds, and the service interruption window is controlled within 3 seconds.

[0085] In this way, multimodal sensor fusion and millisecond-level response mechanism shorten the emergency departure decision time from 8.5 minutes of traditional manual handling to 22 seconds, and reduce the incidence of major mooring accidents by 92%; case knowledge graph construction technology improves the efficiency of historical experience conversion by 76%, and compresses the rule base update cycle from quarterly to hourly; the digital twin verification platform increases the iteration speed of decision path optimization by 40 times, and in the typhoon transit scenario test, the safety margin of the ship dispatch plan is increased by 37%, and the port recovery time is shortened by 58%. These technological advances form a complete technical chain of smart port emergency management and experience evolution, achieving a coordinated leap in port operation safety and operational efficiency.

[0086] In summary, the present invention constructs a ship navigation situation awareness system by integrating multi-source heterogeneous data, significantly improving the real-time and accuracy of port scheduling decisions. The spatiotemporal alignment technology of real-time meteorological and hydrological data effectively eliminates the spatiotemporal dislocation of environmental interference factors, and enhances the characterization ability of the ship's motion state and the coupling relationship between the environment. The ship behavior prediction model based on deep learning realizes the quantitative assessment of multi-dimensional navigation risks, and accurately identifies potential risks such as channel deviation and berthing conflicts. The collaborative optimization mechanism of the dynamic scheduling rule base and risk indicators generates a collaborative decision-making plan that takes into account both safety and efficiency, and optimizes the ship's trajectory control accuracy through dynamic programming of speed adjustment and turning timing. The automated instruction generation and execution system realizes the closed-loop linkage of port resource allocation and ship navigation control, ensures the reliability and response speed of port operation processes in complex operating environments, and improves the overall port navigation efficiency and safety level.

[0087] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the intelligent auxiliary decision-making method for ship entry, exit, berthing and departure based on multi-source data.

[0088] An embodiment of the present invention provides a processor, which is used to run a program, wherein when the program is run, the method for intelligent auxiliary decision-making for ship entry, exit, berthing and departure based on multi-source data is executed.

[0089] In the embodiment of the present invention, Figure 2 As shown, the intelligent auxiliary decision-making system 100 includes at least one processor 101, and at least one memory 102 and a bus 103 connected to the processor 101; wherein the processor 101 and the memory 102 communicate with each other through the bus 103; the processor 101 is used to call the program instructions in the memory 102 to execute the above-mentioned intelligent auxiliary decision-making method for ship entry, exit and berthing based on multi-source data.

[0090] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, intelligent decision-making assistance systems (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] In a typical configuration, the intelligent decision-making support system includes one or more processors (CPU), memory and bus. The intelligent decision-making support system may also include input / output interface, network interface and the like.

[0092] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0093] Computer-readable media include permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage computer-readable storage media or any other non-transmission media that can be used to store information that can be accessed by intelligent decision-making systems. As defined in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carriers.

[0094] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or computer-readable storage medium that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such process, method, commodity or computer-readable storage medium. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the presence of other identical elements in the process, method, commodity or computer-readable storage medium that includes the elements.

[0095] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0096] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. An intelligent auxiliary decision-making method for ship entry and exit and berthing based on multi-source data, characterized in that: The method comprises: Acquire multi-source heterogeneous data in the target port area, wherein the multi-source heterogeneous data includes real-time meteorological data, hydrological monitoring data, ship dynamic trajectory data, port facility status data and historical navigation record data; Performing spatiotemporal alignment and feature fusion on the multi-source heterogeneous data to generate a navigation situation feature set of the target ship, wherein the navigation situation feature set includes ship navigation state parameters, a set of environmental interference factors, and a sequence of port resource occupancy states; Input the navigation situation feature set into a pre-trained ship behavior prediction model, and output a navigation risk index set of the target ship under multiple decision dimensions, wherein the navigation risk index set includes a channel deviation probability, a berthing conflict probability, a navigation path feasibility score, and a port operation efficiency impact coefficient; Based on the navigation risk indicator set and the preset port dynamic scheduling rule base, a collaborative decision-making scheme for the entry, exit and berthing of the target ship is generated, wherein the collaborative decision-making scheme includes a speed adjustment sequence, a turn timing set, a berth allocation priority and a berthing and unberthing operation time window; A ship control instruction set is generated according to the collaborative decision-making scheme, and the ship control instruction set is sent to the navigation equipment and port operation equipment of the target ship through the port dispatching system to dynamically adjust the ship's navigation trajectory and port resource allocation status.

2. The method according to claim 1, characterized in that The step of performing spatiotemporal alignment and feature fusion on the multi-source heterogeneous data to generate a navigation situation feature set of the target ship includes: Extracting the ship position sequence, speed sequence and heading angle sequence from the ship dynamic trajectory data to construct a ship kinematics feature vector; Performing spatial grid matching on the wind speed, wind direction, and visibility in the real-time meteorological data and the tidal height and flow velocity in the hydrological monitoring data to generate an environmental interference factor matrix; Constructing a time series of port resource occupancy status according to the berth occupancy status, loading and unloading equipment working status and waterway congestion index in the port facility status data; The ship kinematic feature vector, environmental interference factor matrix and port resource occupancy state sequence are time-stamp synchronized and spatial coordinate mapped to generate a multi-dimensional fusion navigation situation feature set, in which each feature unit is associated with the real-time position and future predicted time point of the target ship.

3. The method according to claim 2, characterized in that The training process of the ship behavior prediction model includes: Collect historical multi-source heterogeneous data and corresponding actual ship navigation decision records to construct a training sample set, which includes a historical navigation situation feature set and annotated navigation risk indicator true values; Construct a multimodal deep learning network, which includes a spatiotemporal convolution module, an attention weight allocation module, and a dynamic feature interaction module, wherein: The spatiotemporal convolution module is used to extract the spatial correlation between the local motion pattern of the ship trajectory and the environmental interference factors; The attention weight allocation module dynamically adjusts the contribution of meteorological data, hydrological data and port status data to the impact of ship behavior; The dynamic feature interaction module simulates the coupling effect between the ship motion characteristics and the environmental interference factors at continuous time steps; The loss function of the multimodal deep learning network is optimized by a back-propagation algorithm, wherein the loss function includes the mean square error of the channel deviation probability, the cross entropy of the berthing conflict probability, and the ranking loss of the navigation path feasibility score.

4. The method according to claim 3, characterized in that The generating of a collaborative decision-making scheme for the entry, exit and berthing of the target ship comprises: Determine the channel right of way level of the target ship according to the channel deviation probability and the channel priority table in the port dynamic scheduling rule base; Based on the berthing conflict probability and the berth allocation history data, calculating the berth suitability score of the target ship, the score including the matching degree between the ship draft and the berth water depth, and the ratio of the ship length to the length of the berth free area; Combining the navigation path feasibility score with the port operation efficiency impact coefficient, generating a set of constraint conditions for the speed adjustment sequence, wherein the constraint conditions include a minimum safe speed threshold, a maximum turning angle rate, and an environmental interference tolerance; The waterway right level, berth suitability score and speed adjustment constraints are weightedly integrated through a multi-objective optimization algorithm to generate a collaborative decision-making plan that includes time window constraints and resource conflict avoidance.

5. The method according to claim 4, characterized in that The generation and execution of the ship control instruction set includes: Decomposing the speed adjustment sequence in the collaborative decision-making scheme into discrete control instructions, each instruction including a target speed value, an acceleration limit and an effective time interval; generating a steering angle control signal sequence according to the steering opportunity set, wherein the signal sequence is matched with the terrain feature points of the port channel in real time; Mapping the berth allocation priority into a berth guidance instruction, wherein the instruction includes berth coordinates, berthing angle, and activation timing of cable mooring equipment; The ship control instruction set is pushed to the target ship's autopilot system and port operation control terminal in real time through the port dispatching system, and the ship's response status and port equipment feedback data during the execution of the instructions are monitored.

6. The method according to claim 5, characterized in that The method further comprises: Real-time collection of the execution feedback data of the ship control instruction set, including the actual speed deviation of the ship, the steering angle execution error and the change in berth occupancy status; Inputting the execution feedback data into the ship behavior prediction model for online incremental learning, and dynamically adjusting the environmental factor contribution parameters in the attention weight allocation module of the model; Recalculate the navigation risk indicator set based on the updated ship behavior prediction model to generate a revised collaborative decision-making plan; When it is detected that the waterway congestion index exceeds a preset threshold or the probability of berth conflict continues to rise, the real-time reconstruction process of the collaborative decision-making scheme is triggered, and a collaborative collision avoidance instruction is sent to the associated ships.

7. The method according to claim 6, characterized in that The online incremental learning process includes: Extracting the ship response delay time, environmental interference mutation events and equipment failure abnormality codes from the execution feedback data to construct a model error feature vector; Through the sliding time window mechanism, the deviation between historical prediction results and actual navigation data is statistically analyzed to identify the weak links in the stability of model prediction; The knowledge distillation technology is used to take the fully trained ship behavior prediction model as the teacher model to generate the distillation loss function of the lightweight student model. The student model is fine-tuned using the execution feedback data and the distillation loss function, and the fine-tuned model parameters are synchronized to the edge computing node of the port scheduling system.

8. The method according to claim 7, characterized in that The generation and distribution of the collaborative collision avoidance instruction includes: When it is detected that the predicted navigation paths of multiple ships have overlapping areas in the time and space dimensions, the center coordinates of the overlapping areas and the conflict time windows are extracted; Calculate the collision avoidance priority weight of each ship, the weight being based on the ship's tonnage, the cargo hazard level, and the current adjustable speed range; Generate differentiated speed adjustment suggestions and steering angle compensation amounts according to the collision avoidance priority weights, and establish a communication negotiation protocol between ships; The collision avoidance instructions and negotiation protocols are broadcast to associated ships through the ship self-organizing network, and instruction confirmation signals from each ship are received to update the collaborative decision-making plan.

9. The method according to claim 8, characterized in that The method further comprises: During the berthing and unberthing process of the target ship, real-time monitoring of cable tension sensor data, tidal change rate and workload of terminal loading and unloading equipment; When it is detected that the cable tension exceeds the material strength threshold or the tidal change causes insufficient berth water depth, an emergency unberthing instruction set is generated, which includes emergency thruster start-up parameters, cable quick release sequence and port tugboat dispatch priority; The current berthing operation is forcibly interrupted through the port emergency control system, and the automated execution process of the emergency unberthing instruction set is activated.

10. The method according to claim 9, characterized in that The method further comprises: After the ship completes the port entry and exit and berthing and unberthing operations, the decision execution data and external environment monitoring data of the whole process are collected to build a case knowledge graph; Extract key decision nodes, risk handling modes and resource scheduling optimization paths in the case knowledge graph, generate empirical rules and supplement them to the port dynamic scheduling rule base; Simulating decision paths under different environmental interference scenarios through a reinforcement learning algorithm, and updating the weight allocation strategy of the multi-objective optimization algorithm; The optimized decision logic is deployed to the digital twin platform of the port dispatching system to achieve closed-loop verification of historical case data and real-time decision-making processes.

Images