A Digital Twin-Based Approach to Intelligent Lock Scheduling and Lock Sequence Optimization

Abstract

This invention provides a digital twin-based intelligent scheduling and lock sequence optimization method for ship locks, relating to the field of intelligent logistics and operational scheduling management technology. This digital twin-based intelligent scheduling and lock sequence optimization method includes S1: collecting Automatic Identification System (AIS) data, lock operation data, hydrological data, meteorological data, and waterway operation data; constructing a digital twin model of the ship lock; and mapping the status of vessels awaiting lock entry, lock chamber resource status, and waterway traffic status in real time based on the model. By integrating AIS information, lock operation information, hydrological environment information, and waterway operation information, the status of vessels awaiting lock entry, lock chamber occupancy status, and waterway traffic status are synchronously mapped, enabling the scheduling process to have stronger scene correlation capabilities and maintain consistency between the operational status and the actual situation in complex navigation environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics and operation scheduling management technology, specifically to a method for intelligent scheduling and lock sequence optimization based on digital twins. Background Technology

[0002] The field of intelligent logistics and operations scheduling management technology mainly involves the use of information technology, data analysis, and intelligent algorithms to optimize resource allocation, transportation planning, scheduling strategies, and operational efficiency in logistics systems and transportation networks. This field encompasses transportation network planning, warehousing and distribution management, vehicle scheduling and route optimization, workload balancing, dynamic resource allocation, and real-time monitoring. It emphasizes improving the efficiency of logistics systems, reducing operating costs, and enhancing service quality through digital and intelligent means, and achieving scientific decision-making and automated scheduling through data-driven methods. Among these methods, the intelligent scheduling and lock sequence optimization method based on digital twins refers to the establishment of digital twin models of locks and related waterways to simulate and predict ship arrival times, lock chamber resource utilization, and waterway operation status in real time. This is combined with intelligent algorithms to optimize lock sequence scheduling and scheduling strategies, thereby improving lock throughput efficiency, reducing waiting time, optimizing energy consumption, and enhancing the overall operational level of waterway transportation.

[0003] In actual operation, existing lock scheduling methods mostly rely on the current navigation status, with scheduling decisions focusing on immediate conditions and lacking sufficient consideration for subsequent vessel arrivals and channel status evolution. This easily leads to vessel congestion during peak hours. For example, when multiple vessels arrive at the same lock in a short period, the existing schedule cannot handle the increased flow in time, resulting in continuously increasing waiting times and further impacting the operational efficiency of subsequent segments. The current scheduling process focuses more on the sequential arrangement of vessels, with insufficient consideration for the utilization of space within the lock chambers. This results in some lock sessions having low loading rates and idle resources, preventing the full utilization of unit throughput capacity. As multiple locks form continuous navigation links, local congestion can easily spread to adjacent areas. The existing operating methods lack comprehensive analysis of the overall interconnectedness, leading to the continuous accumulation of pressure at some nodes and affecting the operational stability of the entire waterway network. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for intelligent scheduling and lock sequence optimization based on digital twins. This method solves the problem that existing lock scheduling modes, in actual operation, mostly rely on the current navigation status for scheduling, with scheduling decisions focusing on the immediate situation and lacking sufficient consideration for changes in subsequent ship arrivals and the evolution of waterway conditions, which easily leads to concentrated ship gatherings during peak periods.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent scheduling and lock sequence optimization based on digital twins for ship locks, comprising the following steps: S1. Collect Automatic Identification System (AIS) data, lock operation data, hydrological data, meteorological data, and waterway operation data; construct a digital twin model of the lock; and map the status of the vessel to be locked, the lock chamber resource status, and the waterway traffic status in real time based on the model. S2. Establish a ship arrival time prediction model and obtain the predicted arrival time of each ship waiting to enter the lock within a preset time window through the digital twin model. S3. Generate a candidate gate sequence arrangement scheme based on the predicted arrival time and ship attribute information, and optimize the spatial layout based on the three-dimensional spatial loading model of the lock chamber; S4. Construct a multi-objective optimization model, taking ship waiting time, lock chamber utilization rate, ship punctuality rate and lock operation energy consumption as optimization objectives, and determine the optimal lock scheduling strategy through reinforcement learning methods. S5. Establish a lock network topology model, use graph neural networks to analyze node congestion status and dynamically adjust the lock scheduling strategy. S6. Update the digital twin model and scheduling strategy based on actual operation feedback data to achieve continuous optimization of intelligent scheduling of the lock.

[0006] Preferably, the digital twin model of the lock includes a ship twin sub-model, a lock chamber twin sub-model, a waterway twin sub-model, and an environmental twin sub-model; The ship twin model should include at least the ship's length, beam, draft, cargo capacity, speed, and heading parameters, while the lock chamber twin model should include at least the lock chamber dimensions, water level status, and equipment operating status parameters.

[0007] Preferably, the ship arrival time prediction model is constructed using a Long Short-Term Memory (LSTM) network, taking historical AIS trajectory sequences, speed change sequences, heading change sequences, and environmental impact parameters as inputs, and the actual arrival time of the ship as the output variable.

[0008] Preferably, the candidate gate sequence arrangement scheme is generated based on the ship size, cargo attributes, safety level and predicted arrival time to establish ship grouping rules, and the candidate gate sequence corresponding to each ship is determined according to the grouping rules.

[0009] Preferably, the three-dimensional spatial loading model of the lock chamber abstracts the ship as a loading object with spatial constraints, and the lock chamber as a target loading space. The spatial layout of the ship in the lock chamber is determined by the three-dimensional packing algorithm to improve the space utilization rate of a single lock operation.

[0010] Preferably, the three-dimensional spatial loading model is solved using a particle swarm optimization algorithm, setting constraints on ship collision, safety distance, lock chamber boundary, and ship entry and exit sequence, and obtaining the optimal lock allocation scheme based on the constraints.

[0011] Preferably, the multi-objective optimization model is constructed using reinforcement learning methods, taking the ship queuing status, lock chamber operation status, water level status, and equipment status as environmental state variables, and the lock opening decision, ship allocation decision, and priority adjustment decision as action variables, and constructing the average waiting time, lock chamber utilization rate, ship punctuality rate, and unit energy consumption index as a joint reward function.

[0012] Preferably, the reinforcement learning method employs the Proximal Policy Optimization (PPO) algorithm or the Multi-Objective Deep Q-Network (MO-DQN) algorithm, and trains the optimal scheduling strategy through a digital twin model simulation environment.

[0013] Preferably, the lock network topology model includes multiple lock nodes, anchorage nodes, and channel nodes. Graph neural networks are used to analyze the ship flow relationships between nodes, predict the congestion status of each node in the future, and dynamically adjust the lock scheduling strategy accordingly.

[0014] Preferably, the self-learning mechanism of the scheduling strategy calculates the average waiting time of ships, lock chamber utilization rate, scheduling deviation rate and equipment utilization rate based on actual operation feedback data, uses Bayesian optimization algorithm to determine the direction of scheduling parameter update, and updates the ship arrival time prediction model and lock sequence arrangement model through incremental learning to achieve continuous optimization of the lock scheduling strategy.

[0015] This invention provides a method for intelligent scheduling and lock sequence optimization based on digital twins. It has the following beneficial effects: This invention integrates vessel automatic identification information, lock operation information, hydrological environment information, and waterway operation information to synchronously map the status of vessels awaiting lock entry, lock chamber occupancy status, and waterway traffic status. This enhances the scheduling process's ability to correlate scenarios, maintaining consistency between operational status and actual conditions in complex navigation environments. It reduces the impact of deviations from single information sources and, by combining preset time windows for forward analysis of vessel arrivals, shifts lock scheduling from reactive response to advance planning. This helps reduce queuing caused by concentrated arrivals, improves the continuity of traffic organization, generates candidate scheduling results based on vessel attributes, and optimizes loading methods based on spatial layout relationships, ensuring optimal performance within the lock chamber. The space utilization is more balanced, avoiding waste of spare space due to unreasonable loading distribution, improving the carrying capacity per lock cycle, and balancing waiting time, resource utilization level, on-time operation and energy consumption. The scheduling process takes into account both efficiency and economy, avoiding the loss of overall benefits due to local performance improvement caused by a single goal. The correlation between locks is analyzed to determine the trend of congestion spread, and the operation status and scheduling strategy are continuously corrected based on operation feedback. This enables different locks to form dynamic coordination capabilities, enhances the adaptability of complex navigation networks to traffic fluctuations, promotes smoother resource allocation, reduces the spread of local congestion to the overall waterway, and improves the stability and long-term optimization capabilities of lock operation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the main steps of the present invention; Figure 2 This is a schematic diagram illustrating the construction and data acquisition of the digital twin of the ship lock according to the present invention; Figure 3 This is a schematic diagram of the ship arrival time prediction model of the present invention; Figure 4 This is a schematic diagram of the candidate gate arrangement and three-dimensional space optimization of the present invention; Figure 5 This is a schematic diagram of the multi-objective optimization and reinforcement learning scheduling of the present invention; Figure 6 This is a schematic diagram illustrating the dynamic adjustment and self-learning optimization of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: like Figure 1-6As shown, this embodiment of the invention provides a method for intelligent scheduling and lock sequence optimization based on digital twins, including the following steps: S1. Collect Automatic Identification System (AIS) data, lock operation data, hydrological data, meteorological data, and waterway operation data to construct a digital twin model of the lock, and map the status of the vessel to be locked, the lock chamber resource status, and the waterway traffic status in real time based on the model. The collection of Automatic Identification System (AIS) data, lock operation data, hydrological data, meteorological data, and waterway operation data begins with the analysis of the AIS data. This includes the ship's position (latitude and longitude), heading, speed, and ship type information. A time-series decomposition method is then used to calculate the ship's motion vector change every 5 minutes. For example, if a ship's position at 08:00 is... Speed ​​12 knots, 08:05 position ,pass Calculate the longitude movement. Calculate the latitudinal movement to obtain the flight vector. Subsequently, the lock operation data, such as gate opening and closing times, lock chamber water level changes, and lock passage duration, are cleaned. Outliers are defined as records exceeding the mean ± 3σ and interpolated. For example, if the average opening time of a lock chamber is 15 minutes and the standard deviation is 3 minutes, the threshold range is 6-24 minutes. Records exceeding this range are corrected using linear interpolation. Hydrological data, such as flow velocity and direction, are obtained hourly using a raster sampling method. Meteorological data, including wind speed, wind direction, and precipitation, are calculated hourly using a weighted average method. Channel operation data involves the calculation of vessel density and channel capacity. The vessel density per unit area is obtained by dividing the number of vessels by the channel's navigable area. All data are input into a digital twin model. The model maps vessel positions, lock chamber status, and channel traffic status through a spatial grid to achieve real-time mapping.

[0019] S2. Establish a ship arrival time prediction model and obtain the predicted arrival time of each ship waiting to enter the lock within a preset time window through a digital twin model. To establish a ship arrival time prediction model, the current position, speed, heading, and channel hydrological conditions of the ship obtained from the digital twin model are first input into the prediction module. Regression analysis or time series prediction methods are used to calculate the ship arrival time within a preset time window. For example, if a ship is currently 5 kilometers away from the lock and its current speed is 10 knots, with a tidal correction coefficient k=0.9, then the predicted arrival time T=distance / (speed×k)=5km / (10 knots×0.514m / s / knot×0.9)≈1,077 seconds≈18 minutes. Combining the ship type correction factor, such as +2 minutes for cargo ships and -1 minute for passenger ships, the corrected arrival time is obtained. All predicted values ​​are sorted according to the ship arrival sequence, and abnormal deviation values ​​are eliminated using the standard deviation method. The predicted arrival time is then mapped back to the digital twin model to form a dynamic ship arrival status.

[0020] S3. Generate candidate lock sequence arrangement schemes based on predicted arrival time and vessel attribute information, and optimize the spatial layout based on the three-dimensional spatial loading model of the lock chamber; Candidate lock sequence arrangement schemes are generated based on predicted arrival times and vessel attributes. First, vessels are grouped according to arrival time and attributes such as cargo hold capacity, beam, and length. For example, if two cargo ships are 12 meters wide and 110 meters long, and the lock chamber is 25 meters wide and 250 meters long, then the lock chamber space utilization rate is calculated. For each group, candidate gates are generated. Then, using a three-dimensional spatial loading model, the ships are arranged according to their size and the minimum safe distance between them (e.g., 5 meters). The ship positions are adjusted through an iterative algorithm to optimize the spatial layout of the candidate gates. The position coordinates are updated and the new space utilization rate is calculated in each iteration. If the change in space utilization rate is less than 0.01, the iteration stops and the optimal spatial layout scheme is generated.

[0021] S4. Construct a multi-objective optimization model, taking ship waiting time, lock chamber utilization rate, ship punctuality rate and lock operation energy consumption as optimization objectives, and determine the optimal lock scheduling strategy through reinforcement learning methods. A multi-objective optimization model is constructed, taking ship waiting time W, lock chamber utilization rate U, ship punctuality rate P, and lock operation energy consumption E as optimization objectives. First, each objective is normalized; for example, W is normalized to... ,set up =5 minutes =60 minutes, after normalization we get 0.2≤W'≤0.9, then build a reinforcement learning model, take each candidate gate state as the environment state S, select action A as the gate orchestration policy, and define the reward. Weight The calculation formula is obtained by iteratively updating the Q value. With a learning rate η=0.1 and a discount factor γ=0.9, the objective function value is calculated and the optimal action is selected in each iteration, and the optimal gate arrangement strategy is finally output.

[0022] S5. Establish a lock network topology model, use graph neural networks to analyze node congestion status and dynamically adjust the lock scheduling strategy. A lock network topology model is established, where nodes represent lock chambers and edges represent channel connections. A graph neural network (GNN) is used to analyze node congestion. The state vector of each node, including the number of waiting vessels, lock chamber utilization rate, and predicted arrival time of vessels, is input into the GNN to calculate the node congestion index. Set weights The congestion index of each node is compared with a preset threshold of 0.7 based on the number of ships, utilization rate of 0.3, and arrival time deviation of 0.2. If CI>0.7, it is marked as a high-congestion node, and the gate scheduling strategy is adjusted accordingly. By rearranging candidate gates and adjusting ship groups, the ship status coordinates and lock chamber usage order in the digital twin model are updated.

[0023] S6. Update the digital twin model and scheduling strategy based on actual operation feedback data to achieve continuous optimization of intelligent scheduling of the lock.

[0024] The digital twin model and scheduling strategy are updated based on actual operational feedback data. First, actual gate opening and closing times, actual vessel arrival times, and channel hydrological and meteorological data are collected and compared with the predicted values ​​in the digital twin model. The deviation ΔT = actual arrival time - predicted arrival time, and ΔW = actual waiting time - predicted waiting time are then calculated. Finally, the sum of squared errors is calculated. As a model calibration index, the internal state mapping of the digital twin model is updated by adjusting parameters such as the tidal current coefficient k, the ship type correction factor, and the spatial layout safety distance. At the same time, the Q-value table is updated according to the deviation between the historical decisions and the actual situation of the reinforcement learning model. The new strategy output candidate gate sequence is iteratively calculated, and the updated ship status and lock chamber utilization are re-input into the digital twin to achieve continuous updating of the scheduling strategy.

[0025] The digital twin model of the lock includes a ship twin sub-model, a lock chamber twin sub-model, a waterway twin sub-model, and an environmental twin sub-model; The ship twin model should include at least the ship's length, beam, draft, cargo capacity, speed, and heading parameters, while the lock chamber twin model should include at least the lock chamber dimensions, water level status, and equipment operating status parameters.

[0026] First, the sets of hull shape parameters, loading state parameters, motion state parameters, and facility structure parameters are extracted. Each parameter is then categorized and registered according to its acquisition cycle; for example, hull dimensions are divided into "large, medium, and small" ranges, and loading states are divided into "lightly loaded, medium-loaded, and heavily loaded" ranges. Subsequently, correlation calculations are performed on each parameter, and a comprehensive state variable is defined. Where S represents the comprehensive state quantity, Wi represents the weight of the i-th parameter, Pi represents the normalized result of the i-th parameter, and ∑ represents the accumulation process of each parameter. The weight settings are determined by referring to the frequency of occurrence of each parameter in the historical operation record. For example, the weights corresponding to motion state parameters are in the higher range, and the weights corresponding to structural parameters are in the middle range. The normalization process adopts... Where P represents the current parameter value, Indicates the lower limit of the historical range. This represents the upper limit of the historical interval. When a ship is in the medium-load interval and its navigation status is in the stable interval, the corresponding parameters are substituted into the calculation to obtain the comprehensive state quantity. Then, the internal space interval of the facility, the water level interval, and the equipment operation interval are matched accordingly. The matching results are verified item by item and an object association record is formed. For example, when the state quantity of a ship is in the medium-high interval and the state quantity of the facility is in the normal interval, an association mapping record is established between the two, and the correspondence of various parameters is continuously updated. Finally, a complete association information result is formed for subsequent operation process calls.

[0027] The ship arrival time prediction model is constructed using a Long Short-Term Memory (LSTM) network. It takes historical AIS trajectory sequences, speed change sequences, heading change sequences, and environmental impact parameters as inputs and the actual arrival time of the ship as the output variable.

[0028] First, historical trajectory sequences, velocity change sequences, direction change sequences, and external environment sequences are acquired and sorted according to a unified time window. This time window can be divided into short-period, medium-period, and long-period intervals. Then, the changes between adjacent moments are calculated, denoted as ΔV = Vt - Vt-1, where ΔV represents the velocity change, Vt represents the current velocity value, and Vt-1 represents the velocity value at the previous moment. The direction change is calculated using ΔC = Ct - Ct-1, where Ct represents the current direction value, and Ct-1 represents the previous direction value. Finally, environmental impact factors are quantified, and environmental impact quantities are defined as follows: Where E represents the comprehensive environmental impact, Kj represents the environmental weight of the j-th item, and Ej represents the corresponding environmental parameter value. The weight setting is determined based on the frequency of the impact of each environmental factor on the navigation process in the historical samples. If the impact frequency is in a higher range, a higher weight is assigned; if it is in a normal range, a medium weight is assigned. The sequences are input into the time series processing unit in chronological order, and recursive calculations are performed on the data within multiple consecutive periods. During the calculation, the deviation between the current predicted value and the historical reference value is compared. Where D represents the deviation, Tpre represents the prediction result, and Tact represents the actual recorded result. When the deviation is within the preset allowable range, the set of parameters is retained. When the deviation exceeds the allowable range, the weight range of each parameter is readjusted and the calculation is repeated. For example, the environmental impact weight is adjusted from the medium range to the higher range and then recalculated. The process continues iterating until a stable time prediction result is obtained.

[0029] The candidate gate sequence arrangement scheme is generated based on the ship size, cargo attributes, safety level and predicted arrival time to establish ship grouping rules, and the candidate gate sequence corresponding to each ship is determined according to the grouping rules.

[0030] Based on the collected basic attribute information of the objects to be passed, the object identification information, scale information, transport category information, risk classification information, and estimated arrival information are broken down and processed. Scale information is recorded separately according to length, width, and draft ranges. Objects with a length within a preset medium range are classified into the first scale category, and objects with a length within a preset large range are classified into the second scale category. Transport categories are marked according to general cargo, key regulated cargo, and special cargo categories. Risk classification is assigned a level based on historical operation records, the completeness of declared information, and regulatory requirements. Estimated arrival information is obtained by predicting the planned speed and remaining distance, using the formula "estimated arrival time = current..." The process is calculated as "Time + Remaining Distance ÷ Planned Speed", where the remaining distance is the distance from the current location to the target area, and the planned speed is the permitted speed for the corresponding segment. Then, the parameters are correlated and compared. A classification matrix is ​​established based on consistency of scale category, compatibility of cargo category, proximity of risk level, and proximity of arrival time. Objects with arrival time differences within a preset small range are assigned higher correlation weights, while those with arrival time differences within a preset large range are assigned lower correlation weights. The weight settings are determined with reference to the concurrent passage distribution in historical scheduling records. Finally, the objects are classified and sorted according to their comprehensive correlation values, and the corresponding time windows are matched based on the sorting results to form alternative passage sequences associated with each object.

[0031] The three-dimensional spatial loading model of the lock chamber abstracts the ship as a loading object with spatial constraints and the lock chamber as the target loading space. The spatial layout of the ship in the lock chamber is determined by the three-dimensional packing algorithm to improve the space utilization rate of a single lock operation.

[0032] Based on the acquired parameters of the target space and the space of objects to be arranged, the internal dimensional parameters, effective working area parameters, and object outline parameters of the target space are decomposed. The object outline parameters include length, width, and height ranges, while the target space parameters include effective length, effective width, and boundary reserved ranges. Subsequently, a coordinate reference system is established, dividing the target space into multiple continuous regions. The scale parameters of each object are read sequentially, and a layout sequence is generated according to a descending order of size. For each object, its starting and ending coordinate positions are recorded. The space occupancy rate is calculated using the formula "Occupancy Rate = ...". The total volume of objects is calculated by dividing the total volume of the target space by the total volume of the objects. The total volume of objects is calculated by combining the length, width, and height parameters. In instance processing, larger objects can be placed in the boundary area first, and smaller objects can be inserted into the remaining area. The reserved distance between adjacent objects is checked. If the size of the remaining area is lower than the preset capacity range, the object arrangement order is readjusted. The reserved distance threshold is determined according to the regular interval range in the historical operation specifications. Then, the remaining space parameters of each area are continuously updated, and the coordinate information and occupancy information after each adjustment are recorded to finally form the corresponding spatial arrangement result.

[0033] The three-dimensional spatial loading model is solved using the particle swarm optimization algorithm. Constraints such as ship collision, safety distance, lock chamber boundary, and ship entry and exit sequence are set, and the optimal lock allocation scheme is obtained based on the constraints.

[0034] Based on the established constraint parameter set, position constraint parameters, interval constraint parameters, boundary constraint parameters, and sequence constraint parameters are decomposed. Position constraint parameters are obtained using the outer contour coordinates of any two objects. A conflict is determined when the distance between two objects in the length, width, and height directions simultaneously falls within a preset overlap range. Interval constraint parameters are set according to the conventional safe distance range in management specifications. Boundary constraint parameters are set based on the allowable distance range between the target space edge and the outer contour of the object. Sequence constraint parameters establish a correspondence based on the entry sequence number and exit sequence number. Subsequently, each scheme is iteratively verified, and a comprehensive evaluation value is calculated for each scheme. The comprehensive evaluation value can be expressed as... The result is the cumulative result of each evaluation item multiplied by its corresponding weight. The weight parameters are configured with reference to historical scheduling records, running frequency statistics, and management requirements. A higher weight is assigned when an evaluation item is in the priority range, a medium weight when it is in the normal range, and a lower weight when it is in the restricted range. In instance processing, the current position parameters of each object are read first, then the distance parameters between adjacent objects are checked to see if they fall within the allowed range. After that, the object outline is checked to see if it exceeds the boundary range. Then, the first-entering object is checked to see if it corresponds to the first-exit rule. If any constraint is found to be unsatisfactory, the object arrangement position is modified and the evaluation value is recalculated until all constraints are satisfied, and then the corresponding configuration result is output.

[0035] The multi-objective optimization model is constructed using reinforcement learning. It takes the ship queuing status, lock chamber operation status, water level status, and equipment status as environmental state variables, and the lock opening decision, ship allocation decision, and priority adjustment decision as action variables. It constructs the average waiting time, lock chamber utilization rate, ship punctuality rate, and unit energy consumption index as a joint reward function.

[0036] Based on the construction of the scheduling status parameter set under the lock operation scenario, the ship queuing status information, lock chamber operation status information, water level change status information, and electromechanical equipment status information are decomposed and recorded. The ship queuing status is classified and registered according to the range of the number of ships waiting to pass through the lock, the range of average queuing time, and the range of ship type composition. When the number of ships waiting to pass through the lock is in a preset lower range, it is marked as a lightly loaded state; in a preset middle range, it is marked as a normal state; and in a preset higher range, it is marked as a congested state. The lower range can be set as a certain percentage range below the normal design flow, and the higher range can be set as a certain percentage range exceeding the normal design flow. Subsequently, the lock execution record, ship allocation record, and priority adjustment record are read sequentially and converted into executable decision parameters. In the instance processing, a certain time period can be selected where the ship is in a normal state. Using fleets within the designated queuing area as samples, the cumulative waiting time of vessels is statistically analyzed, and the average waiting index is calculated. The average waiting index is obtained as: Average waiting value = Total waiting time ÷ Total number of vessels, where the total waiting time represents the cumulative waiting time of all vessels within the statistical period, and the total number of vessels represents the number of vessels participating in scheduling within the corresponding statistical period. Then, the lock utilization index, on-time arrival index, and unit energy consumption index are calculated separately. Each index is normalized according to a preset weight, and the weight parameters are determined with reference to the fluctuation range of each index in historical scheduling records. For example, the waiting index weight can be in a higher weight range, and the energy consumption index weight can be in a medium weight range. After multiplying each index by its corresponding weight, the results are accumulated to form a comprehensive evaluation value. The evaluation results corresponding to different decision parameters are continuously recorded to form a joint evaluation result for subsequent scheduling decision iterations.

[0037] The reinforcement learning method employs the proximal policy optimization algorithm (PPO) or the multi-objective deep Q-network algorithm (MO-DQN) to train the optimal scheduling policy in a digital twin model simulation environment.

[0038] Based on the establishment of policy learning parameters in the intelligent scheduling training scenario, the state input parameters, decision output parameters, and policy evaluation parameters are decomposed and processed. The state input parameters include ship position interval information, lock chamber occupancy interval information, and equipment operation interval information. First, the simulation operation records are imported in chronological order, and the continuous operation process is divided into multiple training cycles. In each training cycle, the correlation between the current state parameters and the corresponding decision parameters is recorded. In the instance processing, multiple sets of candidate scheduling schemes can be set within a certain training cycle. The average waiting interval, resource occupancy interval, and energy consumption interval corresponding to each scheme are statistically analyzed. Then, they are sorted according to the comprehensive evaluation value, which can be expressed as the cumulative result of the evaluation item and the corresponding coefficient. The evaluation coefficient is set with reference to the fluctuation range in the historical operation samples. If an evaluation item is in the priority interval, a larger coefficient is assigned; if it is in the ordinary interval, a normal coefficient is assigned. Then, the difference between the evaluation results of adjacent training cycles is compared. When the difference is within the preset convergence interval, the current parameter combination is recorded. When the difference exceeds the preset interval, the decision parameter combination is readjusted and training continues. The recorded information of each cycle is continuously updated and the corresponding policy parameter set is generated to form a scheduling policy result that conforms to the current operation state.

[0039] The lock network topology model includes multiple lock nodes, anchorage nodes, and channel nodes. Graph neural networks are used to analyze the ship flow relationships between nodes, predict the congestion status of each node in the future, and dynamically adjust the lock scheduling strategy accordingly.

[0040] Based on the acquisition of node association parameters in the scenario of coordinated operation of lock groups, the parameters of lock nodes, anchorage nodes, and waterway nodes are decomposed and registered, and a node connection relationship table is established. For any node, the information of inbound flow range, outbound flow range, and dwell time range is recorded. Subsequently, the vessel flow records are statistically analyzed according to the connection relationship between nodes. In the instance processing, multiple consecutive operating cycles can be selected as statistical samples to read the number of vessels entering and leaving each node, and the node load value is calculated. The node load value can be obtained by load value = inbound flow ÷ node processing capacity, where inbound flow represents the statistical cycle. The scale of ships entering a node during the period is considered, and the node's processing capacity represents the allowed processing scale of the corresponding node within the same period. When the load value is in a preset low range, it is considered a smooth state; when it is in a medium range, it is considered a normal state; and when it is in a high range, it is considered a congested state. Then, the load value differences between adjacent nodes are compared, the cumulative change in the flow direction is statistically analyzed, and the congestion level threshold is determined by combining historical operation records. The threshold is set with reference to the historical peak period and normal period operation data range. Then, a corresponding correlation matrix is ​​generated according to the load level of each node, and the node status parameters are continuously updated to form the node congestion status result for future operation cycles.

[0041] The self-learning mechanism of the scheduling strategy calculates the average waiting time of ships, lock chamber utilization rate, scheduling deviation rate and equipment utilization rate based on actual operation feedback data. It uses Bayesian optimization algorithm to determine the direction of scheduling parameter update and updates the ship arrival time prediction model and lock sequence arrangement model through incremental learning to achieve continuous optimization of the lock scheduling strategy.

[0042] Based on the collected feedback data from the lock operation, waiting time parameters, utilization parameters, deviation statistics parameters, and equipment operating parameters are broken down and processed. The waiting time parameter is obtained by the difference between the arrival time and the actual passage time of the vessel. The deviation statistics parameter is obtained by the difference between the planned execution time and the actual execution time. Then, the operation records are organized according to a preset statistical period. In the example processing, the average waiting index, utilization index, and deviation ratio index can be calculated separately. The deviation ratio index is obtained by calculating the deviation ratio as the number of deviation events divided by the total number of statistical events. The number of deviation events represents the scale of events exceeding the preset allowable range, and the total number of statistical events represents the scale of all statistical events within the period. Next, a parameter evaluation table is established based on the changing trends of each indicator, and parameter correction coefficients are determined according to the changing patterns in historical operation samples. The correction coefficients are set with reference to the change range of multiple consecutive statistical periods. When the indicator change range is in a stable range, a conventional correction coefficient is used; when the indicator change range is in a fluctuating range, an enhanced correction coefficient is used. Then, the lock arrival time prediction parameters and lock frequency arrangement parameters are updated based on the corrected parameter results. The newly added operation record increments are written into the model sample library. The parameter verification, parameter correction, and sample update processes are continuously executed to form a continuously updated scheduling model result.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling and lock sequence optimization based on digital twins for ship locks, characterized in that, Includes the following steps: S1. Collect Automatic Identification System (AIS) data, lock operation data, hydrological data, meteorological data, and waterway operation data; construct a digital twin model of the lock; and map the status of the vessel to be locked, the lock chamber resource status, and the waterway traffic status in real time based on the model. S2. Establish a ship arrival time prediction model and obtain the predicted arrival time of each ship waiting to enter the lock within a preset time window through the digital twin model. S3. Generate a candidate gate sequence arrangement scheme based on the predicted arrival time and ship attribute information, and optimize the spatial layout based on the three-dimensional spatial loading model of the lock chamber; S4. Construct a multi-objective optimization model, taking ship waiting time, lock chamber utilization rate, ship punctuality rate and lock operation energy consumption as optimization objectives, and determine the optimal lock scheduling strategy through reinforcement learning methods. S5. Establish a lock network topology model, use graph neural networks to analyze node congestion status and dynamically adjust the lock scheduling strategy. S6. Update the digital twin model and scheduling strategy based on actual operation feedback data to achieve continuous optimization of intelligent scheduling of the lock.

2. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The digital twin model of the lock includes a ship twin sub-model, a lock chamber twin sub-model, a waterway twin sub-model, and an environmental twin sub-model; The ship twin model should include at least the ship's length, beam, draft, cargo capacity, speed, and heading parameters, while the lock chamber twin model should include at least the lock chamber dimensions, water level status, and equipment operating status parameters.

3. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The ship arrival time prediction model is constructed using a Long Short-Term Memory (LSTM) network. It takes historical AIS trajectory sequences, speed change sequences, heading change sequences, and environmental impact parameters as inputs and the actual arrival time of the ship as the output variable.

4. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The candidate gate sequence arrangement scheme is generated based on the ship size, cargo attributes, safety level and predicted arrival time to establish ship grouping rules, and the candidate gate sequence corresponding to each ship is determined according to the grouping rules.

5. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 4, characterized in that: The three-dimensional spatial loading model of the lock chamber abstracts the ship as a loading object with spatial constraints and the lock chamber as the target loading space. The spatial layout of the ship in the lock chamber is determined by the three-dimensional packing algorithm to improve the space utilization rate of a single lock operation.

6. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 5, characterized in that: The three-dimensional spatial loading model is solved using the particle swarm optimization algorithm. It sets constraints such as ship collision, safety distance, lock chamber boundary, and ship entry and exit sequence, and obtains the optimal lock allocation scheme based on the constraints.

7. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The multi-objective optimization model is constructed using reinforcement learning. It takes the ship queuing status, lock chamber operation status, water level status, and equipment status as environmental state variables, and the lock opening decision, ship allocation decision, and priority adjustment decision as action variables. It constructs a joint reward function using the average waiting time, lock chamber utilization rate, ship punctuality rate, and unit energy consumption index.

8. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 7, characterized in that: The reinforcement learning method employs the Proximal Policy Optimization (PPO) algorithm or the Multi-Objective Deep Q-Network (MO-DQN) algorithm, and trains the optimal scheduling strategy through a digital twin model simulation environment.

9. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The lock network topology model includes multiple lock nodes, anchorage nodes, and channel nodes. Graph neural networks are used to analyze the ship flow relationships between nodes, predict the congestion status of each node in the future, and dynamically adjust the lock scheduling strategy accordingly.

10. The intelligent scheduling and lock sequence optimization method based on digital twins for ship locks according to claim 1, characterized in that: The self-learning mechanism of the scheduling strategy calculates the average waiting time of ships, lock chamber utilization rate, scheduling deviation rate and equipment utilization rate based on actual operation feedback data. It uses a Bayesian optimization algorithm to determine the direction of scheduling parameter updates and updates the ship arrival time prediction model and lock sequence arrangement model through incremental learning, so as to achieve continuous optimization of the lock scheduling strategy.

Images