Intelligent ship lock joint scheduling method and system for multi-stage waterway

Abstract

The application discloses a multi-stage channel intelligent ship lock joint scheduling method and system, comprising the following steps: S1, collecting multi-source data of multi-stage ship locks and preprocessing; S2, constructing a dynamic sparse Transformer network model, extracting key dependent relationships between ship locks, and outputting node embedding feature vectors; S3, inputting the embedding feature vectors into the black widow optimization algorithm, performing structure guidance, baby killing mechanism and repair generation joint scheduling scheme; S4, scheduling execution according to the joint scheduling scheme, collecting scheduling execution feedback information; S5, updating the attention sparse connection structure in the Transformer model based on the feedback information, and generating an updated structure graph representation; S6, cyclic optimization is performed, and the updated structure graph representation is used to re-perform joint scheduling. The application is characterized in that the intelligentization, self-adaptation and global collaboration ability of the multi-stage channel ship lock joint scheduling are improved, so that the ship passing efficiency and channel resource allocation are significantly optimized.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering technology, and in particular to a method and system for joint scheduling of intelligent ship locks in multi-stage waterways. Background Technology

[0002] In the current transportation and inland waterway shipping system, the scheduling of cascade locks is a typical resource-constrained task scheduling problem, widely used in traffic organization and navigation management in inland waterways such as the Yangtze River and the Grand Canal. Multi-cascade lock structures typically consist of multiple adjacent or series-connected independent locks, with complex scheduling dependencies between them. Scheduling conflicts, resource bottlenecks, or information delays at any stage can lead to system-wide delays, congestion, or resource waste. Therefore, how to scientifically and efficiently coordinate the scheduling of multi-cascade locks while ensuring navigation safety and improving traffic efficiency has become a key challenge in inland waterway transportation systems.

[0003] In existing technologies, scheduling methods for multi-level locks mainly rely on rule-based human experience-based decision-making or passage priority control strategies constructed based on graph models and heuristic algorithms. These methods can provide basic scheduling feasibility under specific scales or static environments, but when faced with challenges such as dynamic ship arrival flows, complex ship structures, and scheduling feedback delays, the following major problems easily arise: First, traditional methods mostly rely on static graphs or rule templates, which cannot dynamically reflect the changing characteristics of passage states between different locks, resulting in a lack of real-time adaptability in the scheduling scheme; Second, they lack structural awareness and fail to fully explore the high-order dependencies between locks in multi-level structures, resulting in low scheduling efficiency and long ship waiting times; Third, scheduling optimization algorithms mostly adopt general heuristic strategies such as genetic algorithms and ant colony algorithms, which suffer from fast local convergence and easy getting trapped in suboptimal solutions, and have insufficient optimization capabilities under multiple constraints; Fourth, most existing scheduling systems lack a systematic feedback learning mechanism for scheduling execution effects, making it difficult to correct the optimization model through scheduling behavior, and the scheduling system is unable to form self-evolution and adaptive optimization capabilities.

[0004] Furthermore, with the rapid development of artificial intelligence technologies, especially deep learning and graph neural networks, some studies have attempted to introduce graph attention mechanisms and graph convolution into lock scheduling modeling to model and analyze the relationships between lock states. However, most of these models are based on static graph structures, lacking adjustable connection structures and dynamic sparse modeling capabilities, making it difficult to address the high dynamics, sparse dependencies, and information asymmetry issues in real-world multi-stage lock systems. On the other hand, although some studies have introduced scheduling feedback information for model training, in practice, it often fails to be deeply integrated with the optimization module, lacking a structural closed-loop linkage mechanism, making it difficult to achieve co-evolution between scheduling schemes, feedback information, and structural learning.

[0005] Therefore, how to provide a method and system for the joint scheduling of intelligent locks for multi-level waterways is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent joint scheduling method and system for multi-level waterways. This invention fully integrates the dynamic sparse Transformer network structure modeling method and the Black Widow optimization algorithm. By constructing a dynamic sparse attention connection structure, it extracts the key dependencies between locks and introduces a structure guidance mechanism and a structure-aware infanticide mechanism to optimize the scheduling population search process. It also adaptively updates the modeling structure based on scheduling execution feedback information. The invention describes in detail the intelligent optimization process for joint scheduling of multi-level locks in complex waterway environments, which has the advantages of fast scheduling response speed, strong structure modeling capability, high adaptability of scheduling strategy, and significant improvement in traffic efficiency.

[0007] The intelligent lock joint scheduling method for multi-stage waterways according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect multi-source data from each lock in the multi-level waterway and preprocess the multi-source data;

[0009] S2. Construct a dynamic sparse Transformer network model, take the preprocessed multi-source data as node input, perform dynamic sparse selection of attention connections through gating mechanism, extract the key dependencies between multi-level locks, generate a sparse structure graph representation, and output the embedded feature vector of each lock node.

[0010] S3. Using the embedded feature vector as the optimization input parameter, initialize the scheduling population, and use the black widow optimization algorithm to optimize the scheduling and generate a multi-stage lock joint scheduling scheme.

[0011] S4. Implement the actual scheduling according to the multi-level lock joint scheduling scheme, and monitor the status feedback information during the scheduling process, including lock congestion, ship delay status and scheduling deviation.

[0012] S5. Based on the state feedback information, update and train the attention sparse connection structure in the dynamic sparse Transformer network model. The updated sparse structure graph representation is used for a new round of scheduling optimization.

[0013] S6. Repeat steps S2 to S5 to dynamically update and optimize the scheduling strategies of each lock in the multi-level waterway, and complete the iterative adjustment and output of the joint scheduling scheme of the multi-level locks.

[0014] Optionally, the multi-source data specifically includes structural information, operating status, historical navigation data, ship arrival time, ship type and load information, which are used to characterize the navigation relationship and dynamic scheduling characteristics between locks and ships in multi-level waterways.

[0015] Optionally, the preprocessing of the multi-source data specifically includes missing value imputation, normalization, and feature encoding, which are used to construct a standardized input feature tensor adapted to the dynamic sparse Transformer network model.

[0016] Optionally, S2 specifically includes:

[0017] S21. Based on the preprocessed multi-source data, each lock is regarded as a node in the graph structure. An input graph containing node feature information and structural connection relationship is constructed and used as the data input of the dynamic sparse Transformer network model.

[0018] S22. Encode the multi-source data input to each node, extract feature vectors including operating status, historical navigation characteristics, and ship queuing information, and initialize the node representation matrix by combining the location encoding information.

[0019] S23. Set up a gating mechanism module in the dynamic sparse Transformer network model, and assign attention connection weights according to the state differences between nodes, historical collaboration frequency and relative navigation priority index to obtain a fully connected attention candidate graph.

[0020] S24. Sparsify the connection edges in the attention candidate graph, retain the connection relationships with connection scores greater than the set threshold, construct a sparse structure graph representation, and remove low-relevance or invalid node connections.

[0021] S25. Input the sparse structure graph representation into the Transformer encoder structure, and perform attention propagation and context information fusion on the effective connections between nodes through the multi-layer encoder module to extract the high-dimensional semantic representation of each node.

[0022] S26. Output the embedded feature vector of each lock node. The embedded feature vector comprehensively represents the node's scheduling weight, navigation status and key dependencies with other nodes in the multi-level structure.

[0023] Optionally, S3 specifically includes:

[0024] S31. The set of embedded feature vectors for lock nodes output by the dynamic sparse Transformer network model is as follows:

[0025] H = {h1,h2,...,h} N};

[0026] Among them, h i Let represent the embedding vector of the i-th lock node, i∈{1,2,...,N}, where N is the number of lock nodes;

[0027] S32. Initialize the scheduling population based on the embedded feature vector set H:

[0028] P = {X} i |i=1,2,...,M};

[0029] Among them, each individual X i ={p i ,t i}, p i Indicates the order of ship passage routes, t i This represents the passage time window, where M is the population size;

[0030] S33. Define the fitness function as F(X). i ):

[0031] F(X i )=α·T(X i )+β·C(X i )+γ·D(X i );

[0032] Wherein, T(X) i Let C(X) be the total waiting time of the scheduling scheme. i ) represents the cost of lock conflict, D(X) i ) represents the scheduling offset, and α, β, and γ are preset non-negative weight coefficients;

[0033] S34, For individuals in the population (X) i ,X j Crossbreeding is performed to generate offspring individuals.

[0034]

[0035] Where, d ij Represents individual X i With X j The degree of difference in scheduling parameters between them, μ1 is the reproduction intensity control coefficient, μ2 is the reproduction disturbance coefficient, sin(μ2·d ij ) represents the function factor used to simulate nonlinear cross-perturbations;

[0036] S35. Regarding the generated offspring individuals Introducing a structure guidance mechanism and constructing a structure fit function

[0037]

[0038] Perform fine-tuning operations to generate structure-guided individuals.

[0039]

[0040] Where η is the structural guiding strength parameter, hp k (m) and hp k (m+1) represent offspring individuals. Embedded feature vectors of two consecutive locks in the middle, Denotes the Euclidean norm. Represents the structure fit function Regarding offspring individuals The gradient;

[0041] S36. Calculate the fitness value for each structure-guided individual. With structural adjustability score The offspring are divided according to the following rules:

[0042] like and These are classified as high-quality offspring and directly retained, among which R thresh The threshold for structural adjustability;

[0043] like and It is then classified as a repairable offspring and enters the repair process;

[0044] like These are then classified as obsolete offspring and deleted.

[0045] in:

[0046]

[0047] S37. Perform local perturbation repair operations on the structural guide individuals classified as repairable offspring to generate repair candidate individuals. Specifically, it includes:

[0048] Perform limited-range exchange operations on the passage order of adjacent vessels in the scheduling path to adjust the path order;

[0049] After completing the route adjustment, the corresponding travel time window t k Introducing a constrained disturbance Δt k Adjusted to a new time window

[0050] The repaired individual is represented as in The adjusted path order;

[0051] For the repaired individual Recalculate fitness value If satisfied Then If an acceptable repaired offspring is retained, it is discarded; otherwise, F is discarded. thresh The fitness threshold;

[0052] S38. Form a candidate set P by combining high-quality offspring, repaired offspring, and the parent with the best fitness in the previous generation. ′ Perform a mutation operation on each of these individuals to generate mutated individuals.

[0053]

[0054] Where λ is the variation factor, rand() is a random variable in the interval [0,1], and X best This refers to the individual with the best fitness function value in the current population.

[0055] S39. According to the fitness function F(X) i For candidate set P ′ Sort all individuals in the population, select the top M individuals with the best fitness to form the next generation population, and update it to P. (g+1) ;

[0056] S310. Repeat steps S34 to S39 until the set number of iterations is reached or the population fitness converges, and output the individual with the best fitness value. This serves as the joint scheduling scheme for the multi-stage ship locks.

[0057] Optionally, S4 specifically includes:

[0058] S41. Receive and output the multi-level lock joint scheduling scheme, including the passage path of each vessel, the opening and closing time window of each lock and the priority order.

[0059] S42. The multi-stage lock joint scheduling scheme is issued to the lock control system and the ship scheduling center, the time-sharing passage control command of each lock is initiated, and the scheduling information is synchronized to each segment node in real time.

[0060] S43. During the actual passage of ships, collect the operating status data of each lock;

[0061] S44. Synchronously record the real-time operating trajectory and status data of ships, and obtain the passage time, waiting time and deviation from the scheduling path of each ship.

[0062] S45. Based on the collected lock operation status data and real-time ship operation trajectory and status data, identify scheduling anomalies that occur during actual operation.

[0063] S46. Extract the state feedback information during the execution of the scheduling scheme, construct a set of scheduling execution deviation indicators, and use them for dynamic sparse Transformer network model updates and structural adjustments.

[0064] Optionally, the lock's operational status data specifically includes lock opening and closing time, queue length, and number of passages per unit time, used to assess the lock's current throughput capacity and operational load status, serving as the basis for scheduling execution monitoring and feedback analysis.

[0065] Optionally, the scheduling anomalies that occur during actual operation include lock congestion, delays, queue jumping, and resource conflicts, which are used to construct state feedback information and guide the structural updates and scheduling optimization adjustments of the dynamic sparse Transformer network model.

[0066] Optionally, S5 specifically includes:

[0067] S51. Receive and organize the output status feedback information, including lock congestion status, ship delay status and scheduling execution deviation.

[0068] S52. Match the status feedback information with the embedded feature vectors used in the scheduling process, label the deviation performance of each lock node in the scheduling execution, and generate a training dataset for structure learning.

[0069] S53. Input the training dataset into the dynamic sparse Transformer network model and perform supervised update training on the gating mechanism and attention connection structure in the dynamic sparse Transformer network model.

[0070] S54. During the training update process, dynamically adjust the connection relationship between each node, retain the edges with attention weight values ​​greater than the preset threshold, delete the edges with attention weight values ​​lower than the threshold, and reconstruct the sparse connection structure.

[0071] S55. Regenerate the sparse structure graph representation based on the updated sparse connection structure and output new node embedding features to represent the structural position and influence of each lock in the current state.

[0072] S56. The updated sparse structure graph representation is used as input to replace the original structure and applied to the next round of joint scheduling optimization of ships and locks.

[0073] The intelligent lock joint scheduling system for multi-stage waterways according to an embodiment of the present invention includes the following modules:

[0074] The data acquisition and preprocessing module collects multi-source data from various locks in the multi-level waterway and preprocesses the multi-source data.

[0075] The structural modeling module is used to construct a dynamic sparse Transformer network model, extract key dependencies between locks, generate a sparse structural graph representation, and output embedded feature vectors.

[0076] The scheduling optimization module is used to initialize the scheduling population and generate a joint scheduling scheme for multi-level locks using the black widow optimization algorithm.

[0077] The structure guidance module is used to introduce structure embedding information during the scheduling optimization process to fine-tune the individual offspring schedulers generated by crossover.

[0078] The anomaly identification module is used to monitor the lock operation status and vessel passage behavior during the scheduling process, and to identify scheduling anomalies including congestion, delays, queue jumping and resource conflicts.

[0079] The feedback training module is used to update and train the attention sparse connection structure of the dynamic sparse Transformer network model based on the scheduling execution feedback information, thereby optimizing the structure awareness capability.

[0080] The decision output module is used to output the optimized scheduling results, including ship passage paths, time windows and lock control instructions, for the scheduling execution terminal to call and issue.

[0081] The beneficial effects of this invention are:

[0082] This invention introduces a dynamic sparse Transformer network and an improved Black Widow optimization algorithm to achieve efficient modeling and intelligent scheduling optimization of complex dependencies between locks in multi-level waterways. It overcomes the problems of weak structure awareness, susceptibility to local optima, and lack of feedback mechanisms in traditional scheduling methods. By constructing a gated sparse connection structure, dynamically extracting navigation relationships between key nodes, and training the system with multi-source data such as vessel traffic status and lock operation characteristics, the invention effectively improves the scheduling model's adaptability to structural changes and enhances the system's robustness in dynamic scheduling environments.

[0083] The structure-guided mechanism and structure-aware infanticide mechanism integrated in this invention enable the preservation of potentially high-quality scheduling schemes during the optimization process, preventing excellent solutions from being mistakenly eliminated and improving the structural consistency and search directionality of individual scheduling entities. Simultaneously, state feedback information is introduced during scheduling execution to incrementally update the network model. By constructing a closed-loop iterative mechanism of scheduling-execution-feedback-optimization, the system possesses the ability to continuously learn and adaptively evolve, solving the problems of difficulty in dynamically correcting models and the disconnect between scheduling strategies and actual operation in existing technologies.

[0084] In summary, this invention not only improves vessel traffic efficiency and lock resource utilization, but also significantly reduces vessel waiting time and lock congestion risks. It possesses technical advantages such as accurate structural modeling, efficient optimization strategies, strong interpretability of scheduling results, and high system stability. It is suitable for the intelligent scheduling needs of multi-stage navigation systems in modern inland waterway transportation and has good engineering feasibility and application value. Attached Figure Description

[0085] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0086] Figure 1 The flowchart shows the intelligent lock joint scheduling method for multi-stage waterways proposed in this invention.

[0087] Figure 2 This is a schematic diagram of the structure of the intelligent ship lock joint scheduling system for multi-level waterways proposed in this invention. Detailed Implementation

[0088] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0089] refer to Figure 1 The intelligent lock joint scheduling method for multi-level waterways includes the following steps:

[0090] S1. Collect multi-source data from each lock in the multi-level waterway and preprocess the multi-source data;

[0091] S2. Construct a dynamic sparse Transformer network model, take the preprocessed multi-source data as node input, perform dynamic sparse selection of attention connections through gating mechanism, extract the key dependencies between multi-level locks, generate a sparse structure graph representation, and output the embedded feature vector of each lock node.

[0092] S3. Using the embedded feature vector as the optimization input parameter, initialize the scheduling population, and use the black widow optimization algorithm to optimize the scheduling and generate a multi-stage lock joint scheduling scheme.

[0093] S4. Implement the actual scheduling according to the multi-level lock joint scheduling scheme, and monitor the status feedback information during the scheduling process, including lock congestion, ship delay status and scheduling deviation.

[0094] S5. Based on the state feedback information, update and train the attention sparse connection structure in the dynamic sparse Transformer network model. The updated sparse structure graph representation is used for a new round of scheduling optimization.

[0095] S6. Repeat steps S2 to S5 to dynamically update and optimize the scheduling strategies of each lock in the multi-level waterway, and complete the iterative adjustment and output of the joint scheduling scheme of the multi-level locks.

[0096] This invention introduces a dynamic sparse Transformer network and an improved Black Widow optimization algorithm into the multi-level waterway lock scheduling process, achieving dynamic modeling of structural relationships between locks and intelligent optimization of joint scheduling schemes, significantly improving the system's perception and adaptability to complex scheduling structures. The sparse connection mechanism accurately extracts the dependencies of key nodes, enabling the model to maintain computational efficiency while possessing good global structural modeling capabilities. The optimization phase introduces a structure-guided and perception-based elimination mechanism, effectively avoiding the local convergence problem present in traditional intelligent algorithms and improving the structural coordination and search directionality of individual scheduling entities. Simultaneously, this invention constructs a complete scheduling execution feedback closed-loop mechanism, which can feed back information on congestion, delays, and path deviations occurring in the actual scheduling state to the model training process, achieving adaptive updates and evolution of the structural graph and ensuring the model's continuous optimization capability in dynamic environments. Through the iterative execution of S2 to S5, the system achieves continuous iterative optimization of the scheduling strategy, making the scheduling scheme more accurate, reasonable, and efficient, ultimately improving ship passage efficiency and reducing resource conflicts, demonstrating significant engineering practical value and promising prospects for widespread application.

[0097] In this embodiment, the multi-source data specifically includes structural information, operating status, historical navigation data, ship arrival time, ship type and load information, which are used to characterize the navigation relationship and dynamic scheduling characteristics between locks and ships in multi-level waterways.

[0098] In this embodiment, the preprocessing of the multi-source data specifically includes missing value imputation, normalization, and feature encoding, which are used to construct a standardized input feature tensor that adapts to the dynamic sparse Transformer network model.

[0099] In this embodiment, S2 specifically includes:

[0100] S21. Based on the preprocessed multi-source data, each lock is regarded as a node in the graph structure. An input graph containing node feature information and structural connection relationship is constructed and used as the data input of the dynamic sparse Transformer network model.

[0101] S22. Encode the multi-source data input to each node, extract feature vectors including operating status, historical navigation characteristics, and ship queuing information, and initialize the node representation matrix by combining the location encoding information.

[0102] S23. Set up a gating mechanism module in the dynamic sparse Transformer network model, and assign attention connection weights according to the state differences between nodes, historical collaboration frequency and relative navigation priority index to obtain a fully connected attention candidate graph.

[0103] S24. Sparsify the connection edges in the attention candidate graph, retain the connection relationships with connection scores greater than the set threshold, construct a sparse structure graph representation, and remove low-relevance or invalid node connections.

[0104] S25. Input the sparse structure graph representation into the Transformer encoder structure, and perform attention propagation and context information fusion on the effective connections between nodes through the multi-layer encoder module to extract the high-dimensional semantic representation of each node.

[0105] S26. Output the embedded feature vector of each lock node. The embedded feature vector comprehensively represents the node's scheduling weight, navigation status and key dependencies with other nodes in the multi-level structure.

[0106] This invention effectively improves the accuracy of modeling structural relationships between locks and enhances the ability to understand scheduling context by refining the modeling process of dynamic sparse Transformer networks. In the node modeling stage, preprocessed multi-source data is constructed as a graph structure input, giving each lock node a clear feature representation and structural position. By introducing multi-dimensional features such as operating status, historical navigation characteristics, and ship queuing information, comprehensive modeling of scheduling influencing factors is achieved. During the attention connection construction process, connection weights are dynamically allocated based on state differences, historical coordination frequency, and navigation priority. A gating mechanism is used for sparsification to ensure that the model retains only valid connections with strong dependencies, reducing interference from invalid information and improving structural representation capabilities. A multi-layer Transformer encoder performs context fusion processing on the sparse graph structure, generating high-dimensional embedding vectors with global scheduling meaning. The final output node embedding feature vectors not only retain individual state features but also incorporate structural dependencies, providing more accurate and interpretable scheduling inputs for subsequent optimization modules. This significantly enhances the system's scheduling modeling accuracy and intelligent perception capabilities under complex multi-tiered structures.

[0107] In this embodiment, S3 specifically includes:

[0108] S31. The set of embedded feature vectors for lock nodes output by the dynamic sparse Transformer network model is as follows:

[0109] H = {h1,h2,...,h} N};

[0110] Among them, h i Let represent the embedding vector of the i-th lock node, i∈{1,2,...,N}, where N is the number of lock nodes;

[0111] S32. Initialize the scheduling population based on the embedded feature vector set H:

[0112] P = {X} i |i=1,2,...,M};

[0113] Among them, each individual X i ={p i ,t i}, p i Indicates the order of ship passage routes, t i This represents the passage time window, where M is the population size;

[0114] S33. Define the fitness function as F(X). i ):

[0115] F(X i )=α·T(X i )+β·C(X i )+γ·D(X i );

[0116] Wherein, T(X) i Let C(X) be the total waiting time of the scheduling scheme. i ) represents the cost of lock conflict, D(X) i ) represents the scheduling offset, and α, β, and γ are preset non-negative weight coefficients;

[0117] S34, For individuals in the population (X) i ,X j Crossbreeding is performed to generate offspring individuals.

[0118]

[0119] Where, d ij Represents individual X i With X j The degree of difference in scheduling parameters between them, μ1 is the reproduction intensity control coefficient, μ2 is the reproduction disturbance coefficient, sin(μ2·d ij ) represents the function factor used to simulate nonlinear cross-perturbations;

[0120] S35. Regarding the generated offspring individuals Introducing a structure guidance mechanism and constructing a structure fit function

[0121]

[0122] Perform fine-tuning operations to generate structure-guided individuals.

[0123]

[0124] Where η is the structural guiding strength parameter, hp k (m) and hp k (m+1) represent offspring individuals. Embedded feature vectors of two consecutive locks in the middle, Denotes the Euclidean norm. Represents the structure fit function Regarding offspring individuals The gradient;

[0125] S36. Calculate the fitness value for each structure-guided individual. With structural adjustability score The offspring are divided according to the following rules:

[0126] like and These are classified as high-quality offspring and directly retained, among which R thresh The threshold for structural adjustability;

[0127] like and It is then classified as a repairable offspring and enters the repair process;

[0128] like These are then classified as obsolete offspring and deleted.

[0129] in:

[0130]

[0131] S37. Perform local perturbation repair operations on the structural guide individuals classified as repairable offspring to generate repair candidate individuals. Specifically, it includes:

[0132] Perform limited-range exchange operations on the passage order of adjacent vessels in the scheduling path to adjust the path order;

[0133] After completing the route adjustment, the corresponding travel time window t k Introducing a constrained disturbance Δt k Adjusted to a new time window

[0134] The repaired individual is represented as in The adjusted path order;

[0135] For the repaired individual Recalculate fitness value If satisfied Then If an acceptable repaired offspring is retained, it is discarded; otherwise, F is discarded. thresh The fitness threshold;

[0136] S38. Form a candidate set P by combining high-quality offspring, repaired offspring, and the parent with the best fitness in the previous generation. ′ Perform a mutation operation on each of these individuals to generate mutated individuals.

[0137]

[0138] Where λ is the variation factor, rand() is a random variable in the interval [0,1], and X best This refers to the individual with the best fitness function value in the current population.

[0139] S39. According to the fitness function F(X) i For candidate set P ′ Sort all individuals in the population, select the top M individuals with the best fitness to form the next generation population, and update it to P. (g+1) ;

[0140] S310. Repeat steps S34 to S39 until the set number of iterations is reached or the population fitness converges, and output the individual with the best fitness value. This serves as the joint scheduling scheme for the multi-stage ship locks.

[0141] This invention addresses the problems of traditional intelligent scheduling algorithms in multi-level lock joint scheduling, such as fast local convergence, lack of structure awareness, and easy elimination of high-quality individuals. It proposes an improved Black Widow optimization algorithm that incorporates structure information guidance and a hierarchical optimization mechanism. By using the embedded feature vector output by the dynamic sparse Transformer model as the optimization input, the algorithm comprehensively reflects the state, navigation weights, and structural relationships of lock nodes, effectively improving the scheduling feasibility of the initial population. A structure guidance mechanism is introduced during the optimization process, applying a structure fitness gradient after offspring reproduction to achieve consistent adjustment between the scheduling scheme and the structure embedding space, avoiding blind search. Furthermore, a structure-aware individual partitioning mechanism is designed, finely dividing offspring into three categories—high-quality, repairable, and eliminated—based on fitness and structural adjustability. This improves the retention rate of high-quality individuals, and introduces local perturbation operations for fine-tuning repair of repairable individuals, enhancing the algorithm's fault tolerance and evolutionary resilience. During the iteration process, a directional mutation strategy based on the current best individual is introduced to enhance the population convergence speed and search diversity. Ultimately, through the deep integration of structural optimization and scheduling evolution, the scheduling optimization method proposed in this invention can generate higher quality and more structurally compatible joint scheduling schemes, significantly improving the overall performance and intelligence level of multi-level lock scheduling.

[0142] In this embodiment, S4 specifically includes:

[0143] S41. Receive and output the multi-level lock joint scheduling scheme, including the passage path of each vessel, the opening and closing time window of each lock and the priority order.

[0144] S42. The multi-stage lock joint scheduling scheme is issued to the lock control system and the ship scheduling center, the time-sharing passage control command of each lock is initiated, and the scheduling information is synchronized to each segment node in real time.

[0145] S43. During the actual passage of ships, collect the operating status data of each lock;

[0146] S44. Synchronously record the real-time operating trajectory and status data of ships, and obtain the passage time, waiting time and deviation from the scheduling path of each ship.

[0147] S45. Based on the collected lock operation status data and real-time ship operation trajectory and status data, identify scheduling anomalies that occur during actual operation.

[0148] S46. Extract the state feedback information during the execution of the scheduling scheme, construct a set of scheduling execution deviation indicators, and use them for dynamic sparse Transformer network model updates and structural adjustments.

[0149] This invention significantly improves the execution accuracy and model adaptability of multi-level lock joint scheduling schemes during actual operation by constructing a closed-loop mechanism for scheduling execution and feedback. During the scheduling execution phase, the system not only distributes the joint scheduling scheme to the lock control and ship scheduling center to achieve time-sharing collaborative control of each lock, but also comprehensively grasps the actual situation of scheduling execution by real-time collection of lock operating status and ship trajectories. By monitoring behavioral data such as passage time, waiting time, and path deviation, the system can promptly identify abnormal events such as congestion, delays, queue jumping, and resource conflicts that may occur during scheduling execution, and extract quantifiable status feedback indicators to construct a set of scheduling execution deviation indicators. This deviation information is not only used for execution monitoring and scheduling response, but can also serve as a high-quality label input to a dynamic sparse Transformer network model, driving the attention connection structure in the model to be updated and optimized, thereby forming a closed-loop mechanism of scheduling-feedback-structure remodeling. This design effectively solves the problem of the disconnect between scheduling execution and optimization modeling in traditional systems, enabling the system to have self-learning and self-correction capabilities, ensuring that the scheduling model can continuously adapt to environmental changes, and improving the system's operational stability, scheduling robustness, and global coordination level.

[0150] In this embodiment, the operational status data of the lock specifically includes the lock opening and closing time, queue length, and number of passages per unit time, which are used to assess the current passage capacity and operational load status of the lock, and serve as the basic data for scheduling execution monitoring and feedback analysis.

[0151] In this embodiment, the scheduling anomalies that occur during actual operation specifically include lock congestion, delays, queue jumping, and resource conflicts, which are used to construct state feedback information and guide the structural updates and scheduling optimization adjustments of the dynamic sparse Transformer network model.

[0152] In this embodiment, S5 specifically includes:

[0153] S51. Receive and organize the output status feedback information, including lock congestion status, ship delay status and scheduling execution deviation.

[0154] S52. Match the status feedback information with the embedded feature vectors used in the scheduling process, label the deviation performance of each lock node in the scheduling execution, and generate a training dataset for structure learning.

[0155] S53. Input the training dataset into the dynamic sparse Transformer network model and perform supervised update training on the gating mechanism and attention connection structure in the dynamic sparse Transformer network model.

[0156] S54. During the training update process, dynamically adjust the connection relationship between each node, retain the edges with attention weight values ​​greater than the preset threshold, delete the edges with attention weight values ​​lower than the threshold, and reconstruct the sparse connection structure.

[0157] S55. Regenerate the sparse structure graph representation based on the updated sparse connection structure and output new node embedding features to represent the structural position and influence of each lock in the current state.

[0158] S56. The updated sparse structure graph representation is used as input to replace the original structure and applied to the next round of joint scheduling optimization of ships and locks.

[0159] This invention achieves adaptive evolution of a dynamic sparse Transformer network model in the joint scheduling of multi-level locks by constructing a state feedback-driven structure update mechanism, significantly improving the model's ability to perceive actual operating states and its structural reconstruction capabilities. During scheduling execution, the system collects and organizes key feedback information such as lock congestion, vessel delays, and scheduling deviations. This information is then precisely mapped using embedded feature vectors generated during the scheduling phase, annotating the execution deviation performance of each lock node and constructing a targeted structure learning training dataset. Based on this, the training data is input into the dynamic sparse Transformer model, and the gating mechanism and attention connection weights are adjusted through supervised training, enabling the model structure to automatically optimize connection strategies based on historical execution feedback. During the update process, connection edges are filtered based on attention weights, retaining only significant dependencies, constructing a new sparse structure graph, and eliminating low-value or redundant connections, further improving the accuracy and compression efficiency of the structure representation. The updated structure graph and embedded feature vectors more realistically reflect the current system operating state and node importance, serving as input to the optimization module and effectively improving the structural matching degree and global rationality of the next round of scheduling schemes, achieving a self-looping iteration of structure-feedback-optimization. This mechanism enhances the model's ability to respond to operational changes, enabling the scheduling system to possess intelligent characteristics such as dynamic evolution, self-optimization, and continuous learning.

[0160] refer to Figure 2 The intelligent lock joint scheduling system for multi-level waterways includes the following modules:

[0161] The data acquisition and preprocessing module collects multi-source data from various locks in the multi-level waterway and preprocesses the multi-source data.

[0162] The structural modeling module is used to construct a dynamic sparse Transformer network model, extract key dependencies between locks, generate a sparse structural graph representation, and output embedded feature vectors.

[0163] The scheduling optimization module is used to initialize the scheduling population and generate a joint scheduling scheme for multi-level locks using the black widow optimization algorithm.

[0164] The structure guidance module is used to introduce structure embedding information during the scheduling optimization process to fine-tune the individual offspring schedulers generated by crossover.

[0165] The anomaly identification module is used to monitor the lock operation status and vessel passage behavior during the scheduling process, and to identify scheduling anomalies including congestion, delays, queue jumping and resource conflicts.

[0166] The feedback training module is used to update and train the attention sparse connection structure of the dynamic sparse Transformer network model based on the scheduling execution feedback information, thereby optimizing the structure awareness capability.

[0167] The decision output module is used to output the optimized scheduling results, including ship passage paths, time windows and lock control instructions, for the scheduling execution terminal to call and issue.

[0168] Example 1:

[0169] To verify the feasibility of this invention in practice, it was applied to a typical multi-level waterway scheduling and management platform. The system was integrated with a real-time vessel trajectory acquisition system, lock operation status sensors, and a historical scheduling database to construct a closed-loop intelligent scheduling system encompassing structural modeling, optimized scheduling, execution feedback, and model updates. In practical application, the system first uses a data acquisition module to obtain basic information for each vessel, such as its type, load capacity, and estimated arrival time. This information is then combined with dynamic operational information such as the opening and closing status of each lock, queue length, and throughput per unit time to generate a standardized input feature tensor.

[0170] Subsequently, the system constructs a dynamic sparse Transformer network model to extract key dependencies between lock nodes and automatically generate a sparse structure graph representation. Under high throughput pressure, the system can dynamically adjust the scheduling strategies of upstream and downstream locks and automatically infer the structural weight changes between nodes, thereby achieving global resource optimization. Combined with an improved Black Widow optimization algorithm, the system evolves the initial scheduling population and adjusts the direction of scheduling paths through a structure-guided mechanism. It also introduces a structure-aware elimination strategy and local repair operations to enhance the adaptability of individual scheduling processes to complex structures.

[0171] During the scheduling execution phase, the system continuously collects actual vessel passage data, delay times, and path deviation records, while simultaneously monitoring the actual operating load of the locks. Based on the feedback information, the system performs structural update training on the dynamic sparse Transformer model, adjusting attention connection weights and connection sparsity to achieve incremental evolution of the structural model, thereby enhancing its versatility and responsiveness under different operating conditions.

[0172] During the one-month trial period, the system dispatched 472 batches of vessels, involving a total of 1,936 vessels passing through, covering multiple waterways and scenarios with varying operational densities. Results showed that, compared to traditional manual dispatching methods, using this system reduced the average vessel waiting time from 68.2 minutes to 42.5 minutes, increasing the lock's throughput capacity by 21.3% per unit time; the average number of lock congestion alarms decreased from 3.1 to 1.4, and the dispatching deviation rate decreased from 9.3% to 2.1%; simultaneously, user satisfaction with dispatching increased from 72.4% to 94.6%. Furthermore, the system's average response time was reduced to 12.4 seconds, demonstrating extremely high dispatching efficiency and model convergence capability.

[0173] Table 1 Comparison of Application Effects of Intelligent Ship Lock Joint Scheduling System

[0174]

[0175]

[0176] As shown in Table 1, the multi-level intelligent joint scheduling system for ship locks proposed in this invention demonstrates significantly superior performance compared to traditional scheduling schemes across multiple key indicators. In particular, it exhibits strong practical application value in improving ship passage efficiency, reducing resource conflict risks, and enhancing system responsiveness. Regarding ship waiting performance, the system effectively alleviates scheduling delays and information inconsistencies between upstream and downstream locks by introducing dynamic sparse structure modeling and the Black Widow optimization scheduling algorithm, combined with structural guidance and feedback mechanisms. This significantly reduces the average waiting time per ship from 68.2 minutes to 42.5 minutes, a reduction of 37.7%. These results demonstrate that the system can quickly make scheduling decisions based on global structural relationships, optimize resource allocation, and improve navigation flow efficiency.

[0177] Regarding lock utilization, the system can accommodate an average of 8.1 vessels per opening and closing operation, a 28.6% improvement compared to traditional methods. This achievement is attributed to the system's unified modeling and group optimization of vessel structure, load, and path conflicts, resulting in a higher density and more balanced vessel scheduling combination. In terms of system congestion early warning and control, by proactively identifying bottleneck nodes and reallocating scheduling windows, the average daily number of lock congestion alarms has decreased from 3.1 to 1.4, a reduction of 54.8%, significantly lowering the probability of conflicts caused by information mismatch or processing delays during scheduling.

[0178] The scheduling plan deviation rate is a core indicator for measuring the consistency between the system's planning accuracy and actual execution. This invention's system continuously corrects the model connection structure through a dynamic feedback learning mechanism, significantly improving the accuracy of the scheduling plan and reducing the deviation rate from 9.3% to 2.1%. Simultaneously, the lock's throughput per unit time also increased from 12.7 vessels / hour to 15.4 vessels / hour, a 21.3% increase, demonstrating the system's potential to achieve higher scheduling throughput under the same physical conditions.

[0179] User satisfaction with the scheduling system increased from 72.4% to 94.6%, reflecting improvements in usability across multiple dimensions, including scheduling rationality, path arrangement, and human-computer interaction response. Particularly noteworthy is the system's ability to maintain interpretable scheduling results and transparent adjustment processes even under high load conditions. Furthermore, in terms of scheduling response time, the system can complete a full scheduling response within 12.4 seconds, compared to the average response time of 38.5 seconds for traditional manual or static rule systems. This significantly enhances the system's practicality and emergency response capabilities, representing a response speed improvement of over 67%.

[0180] In summary, this data comparison table comprehensively reflects the performance advantages and engineering adaptability of the system of the present invention in the multi-level waterway scheduling problem, verifies the significant technical effects of the present invention in improving ship scheduling efficiency, reducing traffic conflicts, and enhancing structural self-adaptability, and provides a feasible and scalable solution for scheduling optimization problems in complex multi-node traffic scenarios.

[0181] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for joint scheduling of intelligent ship locks in multi-stage waterways, characterized in that, Includes the following steps: S1. Collect multi-source data from each lock in the multi-level waterway and preprocess the multi-source data; S2. Construct a dynamic sparse Transformer network model, take the preprocessed multi-source data as node input, perform dynamic sparse selection of attention connections through gating mechanism, extract the key dependencies between multi-level locks, generate a sparse structure graph representation, and output the embedded feature vector of each lock node. S3. Using the embedded feature vector as the optimization input parameter, initialize the scheduling population, and use the black widow optimization algorithm to optimize the scheduling and generate a multi-stage lock joint scheduling scheme. S4. Implement the actual scheduling according to the multi-level lock joint scheduling scheme, and monitor the status feedback information during the scheduling process, including lock congestion, ship delay status and scheduling deviation. S5. Based on the state feedback information, update and train the attention sparse connection structure in the dynamic sparse Transformer network model. The updated sparse structure graph representation is used for a new round of scheduling optimization. S6. Repeat steps S2 to S5 to dynamically update and optimize the scheduling strategies of each lock in the multi-level waterway, and complete the iterative adjustment and output of the joint scheduling scheme of the multi-level locks. S3 specifically includes: S31. The set of embedded feature vectors for lock nodes output by the dynamic sparse Transformer network model is as follows: ； in, Indicates the first Embedding vectors of each lock node , This refers to the number of lock nodes. S32, Based on the embedded feature vector set Initialize the scheduling population: ； Among them, each individual , Indicates the order of ship passage routes. Indicates the passage time window. Population size; S33. Define the fitness function as follows: : ； in, The total waiting time for the scheduling scheme. The cost of the lock conflict For scheduling offset, , , Preset non-negative weighting coefficients; S34. Regarding individuals in the population Perform crossbreeding to generate offspring individuals. : ； in, Represents an individual and The degree of difference in scheduling parameters between them This is the reproductive intensity control coefficient. This is the reproductive disturbance coefficient. This represents the function factor used to simulate nonlinear cross-perturbations; S35. Regarding the generated offspring individuals Introducing a structure guidance mechanism and constructing a structure fit function : ； Perform fine-tuning operations to generate structure-guided individuals. : ； in, For structural guiding strength parameters, and Representing offspring individuals Embedded feature vectors of two consecutive locks in the middle, Denotes the Euclidean norm. Represents the structure fit function Regarding offspring individuals The gradient; S36. Calculate the fitness value for each structure-guided individual. With structural adjustability score The offspring are divided according to the following rules: like and Those that are classified as high-quality offspring will be directly retained. The threshold for structural adjustability; like and If so, it is classified as a repairable offspring and enters the repair process; like If so, it is classified as a discarded offspring and deleted; in: ； S37. Perform local perturbation repair operations on the structural guide individuals classified as repairable offspring to generate repair candidate individuals. Specifically, it includes: Perform limited-range exchange operations on the passage order of adjacent vessels in the scheduling path to adjust the path order; After completing the route adjustment, the corresponding travel time window will be... Introducing constrained disturbances Adjusted to a new time window ; The repaired individual is represented as ,in The adjusted path order; For the repaired individual Recalculate fitness value If satisfied Then If a repaired offspring is retained, it is discarded; otherwise, it is discarded. The fitness threshold; S38. Form a candidate set by combining high-quality offspring, repaired offspring, and the parent with the best fitness in the previous generation. Perform a mutation operation on each of these individuals to generate mutated individuals. : ； in, As a variable factor, for Interval random variable, This refers to the individual with the best fitness function value in the current population. S39. Based on the fitness function For candidate set Sort all individuals and select the top ones. The individuals with the best fitness form the next generation population, which is then updated to include... ; S310. Repeat steps S34 to S39 until the set number of iterations is reached or the population fitness converges, and output the individual with the best fitness value. This serves as the joint scheduling scheme for the multi-stage ship locks.

2. The intelligent lock joint scheduling method for multi-stage waterways according to claim 1, characterized in that, The multi-source data specifically includes structural information, operational status, historical navigation data, ship arrival time, ship type, and load information, which are used to characterize the navigation relationship and dynamic scheduling characteristics between locks and ships in multi-level waterways.

3. The intelligent lock joint scheduling method for multi-stage waterways according to claim 1, characterized in that, The preprocessing of the multi-source data specifically includes missing value imputation, normalization, and feature encoding, which are used to construct a standardized input feature tensor adapted to the dynamic sparse Transformer network model.

4. The intelligent lock joint scheduling method for multi-stage waterways according to claim 1, characterized in that, S2 specifically includes: S21. Based on the preprocessed multi-source data, each lock is regarded as a node in the graph structure. An input graph containing node feature information and structural connection relationship is constructed and used as the data input of the dynamic sparse Transformer network model. S22. Encode the multi-source data input to each node, extract feature vectors including operating status, historical navigation characteristics, and ship queuing information, and initialize the node representation matrix by combining the location encoding information. S23. Set up a gating mechanism module in the dynamic sparse Transformer network model, and assign attention connection weights according to the state differences between nodes, historical collaboration frequency and relative navigation priority index to obtain a fully connected attention candidate graph. S24. Sparsify the connection edges in the attention candidate graph, retain the connection relationships with connection scores greater than the set threshold, construct a sparse structure graph representation, and remove low-relevance or invalid node connections. S25. Input the sparse structure graph representation into the Transformer encoder structure, and perform attention propagation and context information fusion on the effective connections between nodes through the multi-layer encoder module to extract the high-dimensional semantic representation of each node. S26. Output the embedded feature vector of each lock node. The embedded feature vector comprehensively represents the node's scheduling weight, navigation status and key dependencies with other nodes in the multi-level structure.

5. The intelligent lock joint scheduling method for multi-stage waterways according to claim 1, characterized in that, S4 specifically includes: S41. Receive and output the multi-level lock joint scheduling scheme, including the passage path of each vessel, the opening and closing time window of each lock and the priority order. S42. The multi-stage lock joint scheduling scheme is issued to the lock control system and the ship scheduling center, the time-sharing passage control command of each lock is initiated, and the scheduling information is synchronized to each segment node in real time. S43. During the actual passage of ships, collect the operating status data of each lock; S44. Synchronously record the real-time operating trajectory and status data of ships, and obtain the passage time, waiting time and deviation from the scheduling path of each ship. S45. Based on the collected lock operation status data and real-time ship operation trajectory and status data, identify scheduling anomalies that occur during actual operation. S46. Extract the state feedback information during the execution of the scheduling scheme, construct a set of scheduling execution deviation indicators, and use them for dynamic sparse Transformer network model updates and structural adjustments.

6. The intelligent lock joint scheduling method for multi-stage waterways according to claim 5, characterized in that, The operational status data of the lock specifically includes lock opening and closing time, queue length, and number of passages per unit time. These data are used to assess the current throughput capacity and operational load status of the lock, serving as the basis for scheduling execution monitoring and feedback analysis.

7. The intelligent lock joint scheduling method for multi-stage waterways according to claim 5, characterized in that, The scheduling anomalies that occur in actual operation include lock congestion, delays, queue jumping, and resource conflicts, which are used to construct state feedback information and guide the structural updates and scheduling optimization adjustments of the dynamic sparse Transformer network model.

8. The intelligent lock joint scheduling method for multi-stage waterways according to claim 1, characterized in that, S5 specifically includes: S51. Receive and organize the output status feedback information, including lock congestion status, ship delay status and scheduling execution deviation. S52. Match the status feedback information with the embedded feature vectors used in the scheduling process, label the deviation performance of each lock node in the scheduling execution, and generate a training dataset for structure learning. S53. Input the training dataset into the dynamic sparse Transformer network model and perform supervised update training on the gating mechanism and attention connection structure in the dynamic sparse Transformer network model. S54. During the training update process, dynamically adjust the connection relationship between each node, retain the edges with attention weight values ​​greater than the preset threshold, delete the edges with attention weight values ​​lower than the threshold, and reconstruct the sparse connection structure. S55. Regenerate the sparse structure graph representation based on the updated sparse connection structure and output new node embedding features to represent the structural position and influence of each lock in the current state. S56. The updated sparse structure graph representation is used as input to replace the original structure and applied to the next round of joint scheduling optimization of ships and locks.

9. A joint scheduling system for intelligent locks in multi-level waterways, comprising executing the joint scheduling method for intelligent locks in multi-level waterways as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition and preprocessing module collects multi-source data from various locks in the multi-level waterway and preprocesses the multi-source data. The structural modeling module is used to construct a dynamic sparse Transformer network model, extract key dependencies between locks, generate a sparse structural graph representation, and output embedded feature vectors. The scheduling optimization module is used to initialize the scheduling population and generate a joint scheduling scheme for multi-level locks using the black widow optimization algorithm. The structure guidance module is used to introduce structure embedding information during the scheduling optimization process to fine-tune the individual offspring schedulers generated by crossover. The anomaly identification module is used to monitor the lock operation status and vessel passage behavior during the scheduling process, and to identify scheduling anomalies including congestion, delays, queue jumping and resource conflicts. The feedback training module is used to update and train the attention sparse connection structure of the dynamic sparse Transformer network model based on the scheduling execution feedback information, thereby optimizing the structure awareness capability. The decision output module is used to output the optimized scheduling results, including ship passage paths, time windows and lock control instructions, for the scheduling execution terminal to call and issue.

Images