A method and system for constructing and fusing a port intelligent water transportation knowledge graph
By constructing a smart water transport knowledge graph for ports, the fusion of multi-source data and the explicitness of scheduling rules were realized. Combined with reinforcement learning models, interpretable scheduling schemes were generated, which solved the problems of difficult data fusion, implicit decision-making, and insufficient global optimization in port scheduling, and improved the intelligence and efficiency of port scheduling.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF TECH
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-24
AI Technical Summary
Existing port scheduling technologies suffer from problems such as difficulty in integrating multi-source data, implicit scheduling knowledge, insufficient global optimization decision-making capabilities, and a lack of interpretability in scheduling results.
A knowledge graph for intelligent water transport in ports is constructed. A basic knowledge graph is formed through entity extraction, attribute binding, and relationship construction. The constraint rules of port scheduling are integrated and attached to the graph. Feasibility screening is carried out using graph matching algorithms and constraint propagation mechanisms. Evaluation values are generated by combining reinforcement learning models, and finally, an interpretable scheduling scheme is generated.
It has achieved the fusion of multi-source data and the explicit expression of scheduling knowledge, improved the intelligence level and decision-making efficiency of port scheduling, solved the limitations of manual scheduling mode and the problem of delayed response to dynamic events, and improved the utilization rate of shoreline resources.
Smart Images
Figure CN122453035A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart port technology, and in particular to a method and system for constructing and integrating a smart water transport knowledge graph for ports. Background Technology
[0002] Port scheduling and management involves multiple aspects, including ship berthing, berth allocation, loading and unloading operations, and departure arrangements. The quality of its decisions directly impacts port operational efficiency. Existing port scheduling technologies mainly suffer from the following problems: First, there is a lack of effective integration of multi-source heterogeneous data. The data required for port scheduling includes Automatic Identification System (AIS) data, tidal hydrological data, berth status data, cargo information, equipment status data, etc. These data come from different sources, have different formats, and are updated at different frequencies. They are scattered across multiple independent systems such as maritime affairs, port affairs, shipping agency, and terminal operations. The lack of a unified data model and correlation mechanism makes it impossible for scheduling decisions to obtain comprehensive information.
[0003] Second, scheduling rules are difficult to express explicitly. Port scheduling relies on a large number of specialized rules, including rules for matching ship tonnage with berth capacity, rules for matching ship draft with effective water depth, safety distance requirements for dangerous goods vessels, berthing priority rules for different types of vessels, and rules for utilizing tidal windows. These rules currently exist in unstructured form in the experience of scheduling personnel or in scattered documents, lacking a unified framework for expression, making it difficult to systematically apply and continuously optimize the rules.
[0004] Third, scheduling decisions lack global optimization capabilities. Currently, ports generally adopt a manual scheduling model based on the "first-come, first-served" principle, with dispatchers manually creating berth plans based on experience. When multiple vessels arrive at the port simultaneously or when dynamic events such as weather changes or equipment failures occur, manual adjustments suffer from response lags, making it difficult to achieve global optimization in both time and space dimensions. For continuous berth layouts, manual scheduling is often limited to local adjustments, resulting in low shoreline resource utilization and long vessel waiting times in port.
[0005] Fourth, the scheduling results lack interpretability. Recent research has attempted to apply artificial intelligence models such as deep reinforcement learning to port scheduling, but the internal decision-making logic of these models is opaque, and the output results are difficult for schedulers to understand and verify. This leads to a lack of trust in the intelligent system among schedulers, limiting the practical application of intelligent scheduling technology.
[0006] Therefore, there is an urgent need for a method for constructing and integrating a smart water transport knowledge graph that can integrate multi-source heterogeneous data, structure scheduling rules, achieve global optimization decisions, and is interpretable. Summary of the Invention
[0007] The purpose of this invention is to provide a method and system for constructing and integrating a port smart waterway knowledge graph, in order to solve the problems existing in the current port scheduling technology, such as the difficulty in integrating multi-source data, the implicit nature of scheduling knowledge, insufficient global optimization decision-making capabilities, and the lack of interpretability of scheduling results.
[0008] To achieve the above objectives, this invention provides a method for constructing and integrating a port intelligent waterway knowledge graph, comprising the following steps: Step S1: Construct a basic knowledge graph of port and waterway transportation; Step S2: Integrate the port scheduling constraint rules into the basic knowledge graph to form a hierarchical knowledge graph; Step S3: Based on the constraint rules in the hierarchical knowledge graph, perform feasibility screening on the ships to be scheduled and generate a feasible candidate set; Step S4: Based on the feasible candidate set, generate an evaluation value for each candidate solution to obtain the final evaluation value of each candidate solution; Step S5: Determine whether the final evaluation value meets the requirements. If it does not meet the requirements, adjust the constraint rules and return to step S3. If it meets the requirements, proceed to step S6. Step S6: Select the optimal solution based on the final evaluation value of each candidate solution, and generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining the general large model.
[0009] Preferably, step S1 specifically includes the following steps: Step S11: Entity extraction. Extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. Ship entities include ship name and IMO number identification information; berth entities include berth name and port area identification information. Step S12: Attribute binding, binding key attributes to each entity; the attributes of the ship entity include the overall length of the ship. Maximum draft , load capacity tons and ship type ,in For ship indexing, indicating the first A vessel; attributes of the berth entity include berth length. Berth depth berth tonnage and berth function ,in For berth index, indicating the berth One berth; Step S13: Relationship construction, establishing semantic relationships between entities, including berthing relationships from ships to berths, transportation relationships from ships to cargo, and adjacency relationships from berths to adjacent berths.
[0010] Preferably, the constraint rules in step S2 include hard constraint rules and soft constraint rules; the rules in the rule layer are encoded into constraint vectors by a graph neural network and embedded into the attribute representation of the corresponding entity.
[0011] Preferably, hard constraint rules are expressed in predicate logic form, specifically including: Hard constraint H1: Matching constraint between ship tonnage and berth tonnage, expressed as... ; Hard constraint H2: Matching constraint between vessel length and shoreline length, for a set of vessels continuously berthed within the same shoreline. Satisfying the formula ,in For safety reasons, The total length of the shoreline; Hard constraint H3: Matching constraint between ship draft and effective water depth, denoted as... ,in For safety margin coefficient, For a moment The height of the tidal rise; Hard constraint H4: Matching constraint between ship type and berth function, represented by berth function. Must belong to the ship type The required set of job qualifications; Hard constraint H5: Tidal window timing constraint, ship berthing time Mooring and departure times It must be located within the tidal window, among which For ships The moment of berthing begins For ships The duration of operation; the tidal window is defined as a continuous period of time. ,in For ships The minimum safe tidal rise requirement.
[0012] Preferably, soft constraint rules include: Soft constraint R1: Ship priority scheduling rules, based on the ship's operating type. Determine priority score The priority order is: foreign trade liner ships are the highest, followed by domestic trade liner ships, then feeder barges, then foreign trade barges are the fourth highest, and domestic trade barges are the lowest. Soft constraint R2: Operation time rationality rule, based on planned berthing time. Dock time with human experience The absolute value of the deviation is used as the basis for the reasonableness score; Soft constraint R3: Berth resource utilization maximization rule, based on berth space-time utilization rate. As the basis for scoring, the calculation formula is: ,in The total length of the berth shoreline. This represents the total duration of the statistical period.
[0013] Preferably, in step S3, when performing feasibility screening based on constraint rules, a graph matching algorithm is used, combined with a constraint propagation mechanism, specifically including: Step S31: Represent the attributes of the vessel to be scheduled and the attributes of all candidate berths as attribute subgraphs in a hierarchical knowledge graph; Step S32: Perform graph homomorphic matching to find berth nodes that match the ship attribute subgraph. The matching condition is that hard constraints H1, H2, H3, H4 and H5 are all met simultaneously. Step S33: During the matching process, constraint propagation is performed using the adjacency relationships of the knowledge graph: when the berth... When a berth is marked as occupied, the adjacent berths are automatically updated based on their adjacency relationships. and The available shoreline length is used to write the updated attributes back to the hierarchical knowledge graph. Step S34: Output all berth nodes that meet the matching conditions as a feasible berth candidate set. This set contains all berth indices that satisfy all hard constraints. .
[0014] Preferably, step S4 specifically includes the following steps: Step S41: Generate multiple independent evaluation values for each candidate solution. Calculate priority scores respectively Time rationality score and resource utilization rate score Priority scoring Equal to the priority score of the ship Time rationality score ,in Resource utilization rate score is a scale parameter. Equal to berth space utilization rate ; Step S42: Construct the feature vectors of candidate solutions. For candidate solutions... Its eigenvectors ; Step S43: Construct the state space of the reinforcement learning model. Includes the current port congestion index Number of ships to be dispatched The height of the water level during the next tidal window and equipment availability ; Step S44: Forward inference of the reinforcement learning model, concatenating the vectors. Input a reinforcement learning model, which employs a deep Q-network or policy gradient method, to maximize long-term cumulative reward. To optimize the objective, among which As a discount factor, The return is a single-step return and is defined as the increase in port throughput minus the penalty for ship delays. The model outputs the final evaluation value, which represents the total number of time steps. ; Step S45: Repeat step S44 for each candidate solution to obtain the final evaluation value of all candidate solutions.
[0015] Preferably, the specific method for determining whether the final evaluation value meets the requirements in step S5 is as follows: preset threshold. Calculate the maximum value of the final evaluation of all candidate solutions. ,in Indicates the index of all candidate solutions. Take the maximum value; when When it is determined that the requirement is not met, The condition is then determined to meet the requirements; the specific method for adjusting the constraint rules is as follows: relax one non-critical hard constraint rule, which includes the safety factor of hard constraint H2. Safety margin factor of hard constraint H3 The relaxation method is to increase the safety factor. The value is updated to Safety margin factor The value is updated to .
[0016] Preferably, the specific steps for generating natural language explanations using the general large model in step S6 are as follows: Step S61: Retrieve hard constraint rule instances corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the conditions for satisfying each hard constraint, and generate the first explanatory text. The format of the text is: the ship name satisfies the tonnage requirements, draft requirements, length requirements, tidal window timing constraints and type matching requirements of the berth name. Step S62: Retrieve the soft constraint rules and the values of each evaluation value corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the final evaluation value output by the reinforcement learning model, and generate a second explanatory text, which includes priority score, time rationality score, resource utilization score and the final evaluation value output by the reinforcement learning model. Step S63: Combine the first and second explanation texts, and add the reasons for the final decision to generate a complete natural language explanation; Step S64: Output the natural language interpretation and scheduling scheme together to the human-computer interaction interface.
[0017] This invention also provides a system for constructing and fusing a port smart waterway knowledge graph, used to implement the aforementioned method for constructing and fusing a port smart waterway knowledge graph, including: The knowledge graph construction module is used to build a basic knowledge graph for port and waterway transportation. It includes an entity extraction unit, an attribute binding unit, and a relationship construction unit. The entity extraction unit is used to extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. The attribute binding unit is used to bind key attributes to each entity. The relationship construction unit is used to establish semantic relationships between entities. The hierarchical fusion module is used to integrate port scheduling constraint rules onto the basic knowledge graph to form a hierarchical knowledge graph; the constraint rules include hard constraint rules and soft constraint rules; The feasibility screening module is used to screen the feasibility of ships to be scheduled based on the constraint rules in the hierarchical knowledge graph and generate a set of feasible candidates. The feasibility screening module adopts a graph matching algorithm and combines it with a constraint propagation mechanism. It uses the adjacency relationship of the knowledge graph to propagate constraints. When a berth is marked as occupied, it automatically updates the available shoreline length of adjacent berths. The evaluation value generation module is used to generate an evaluation value for each candidate solution based on the feasible candidate set, and obtain the final evaluation value of each candidate solution. The evaluation value generation module includes an independent evaluation value calculation unit, a feature vector construction unit, a state space construction unit, and a reinforcement learning inference unit. The judgment and adjustment module is used to determine whether the final evaluation value meets the requirements. If it does not meet the requirements, the constraint rules are adjusted and the feasibility screening module is triggered to regenerate the feasible candidate set. If it meets the requirements, the optimal solution selection module is triggered. The optimal solution selection module is used to select the optimal solution based on the final evaluation value of each candidate solution, and to generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining a general large model.
[0018] Therefore, the present invention employs the above-mentioned method and system for constructing and integrating a port intelligent waterway knowledge graph, and the beneficial technical effects are as follows: (1) This invention constructs a basic knowledge graph of port and water transport and integrates hard constraint rules and soft constraint rules into the graph in the form of a rule layer, thereby realizing the explicit expression and structured storage of port scheduling knowledge. This solves the problems in the prior art where it is difficult to integrate multi-source heterogeneous data and the implicit nature of scheduling knowledge makes it difficult to inherit and reuse experience.
[0019] (2) This invention generates a feasible candidate set by performing feasibility screening based on hard constraint rules, and then generates an evaluation value for each candidate scheme by combining soft constraint rules with a reinforcement learning model. When the evaluation value does not meet the requirements, the constraint rules are adjusted and the selection is re-screened through a feedback iteration mechanism. This realizes the optimal allocation of spatiotemporal resources from a global perspective and solves the problems of the existing technology where the manual scheduling mode is difficult to cope with dynamic events and the shoreline resource utilization rate is low.
[0020] (3) This invention retrieves relevant information from a hierarchical knowledge graph using a general large model, generates an optimal scheduling scheme that includes berth allocation, time window and natural language interpretation, and presents the scheduling decision process in natural language form, which solves the problem that most intelligent scheduling models in the prior art are "black box" structures and the decision process is difficult for scheduling personnel to understand and trust. Attached Figure Description
[0021] Figure 1 This is a flowchart illustrating the construction and fusion method of a port intelligent waterway knowledge graph according to the present invention; Figure 2 This is a flowchart for the feasibility screening process; Figure 3 A flowchart is generated to evaluate candidate solutions. Detailed Implementation
[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.
[0024] Example 1 This embodiment uses the daily ship scheduling of a large coastal port as an example to provide a detailed explanation of the construction and fusion method of a port intelligent waterway knowledge graph provided by the present invention.
[0025] The port dispatch center received information on three vessels awaiting dispatch at 8:00 AM that day, as follows: First Vessel: Overall length 180m, maximum draft 10.5m, deadweight tonnage 50,000 tons, vessel type: container ship, operating type: foreign trade liner, estimated operation time: 8 hours, minimum safe tidal rise requirement. =0.8m.
[0026] Second vessel: Overall length 120m, maximum draft 6.2m, deadweight tonnage 20,000 tons, vessel type: bulk carrier, operating type: domestic trade liner, estimated operation time: 6 hours, minimum safe tidal increase requirement. =0.5m.
[0027] The third vessel: 95m in overall length, 5.0m in maximum draft, 10,000 deadweight tons, bulk carrier, feeder barge operation type, estimated operation time of 4 hours, minimum safe tidal rise requirement. =0.3m.
[0028] Current port status: Port congestion index =0.3, Number of ships to be dispatched =3, the next tidal window water level rise is 1.2m, equipment availability rate =0.95. Preset threshold. =0.7, initial value of safety factor =1.2, initial value of safety margin factor =0.9.
[0029] The total length of the shoreline of a certain port area is 1533m, of which the continuous berth section is 1533m long. The water depth of each berth is 15.1m, and the carrying capacity ranges from 70,000 tons to 120,000 tons.
[0030] Reference Figures 1-3 The specific steps in this embodiment are as follows: Step S1: Construct a basic knowledge graph of port and waterway transportation.
[0031] Step S11: Entity extraction. Extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. Ship entities include ship name and IMO number identification information. Berth entities include berth name and port area identification information.
[0032] Step S12: Attribute binding, binding key attributes to each entity; the attributes of the ship entity include the overall length of the ship. Maximum draft , load capacity tons and ship type ,in For ship indexing, indicating the first A vessel; attributes of the berth entity include berth length. Berth depth berth tonnage and berth function ,in For berth index, indicating the berth One berth.
[0033] Step S13: Relationship construction, establishing semantic relationships between entities, including berthing relationships from ships to berths, transportation relationships from ships to cargo, and adjacency relationships from berths to adjacent berths.
[0034] Step S2: Integrate the port scheduling constraint rules into the basic knowledge graph to form a hierarchical knowledge graph.
[0035] Constraint rules include hard constraint rules and soft constraint rules; the rules in the rule layer are encoded into constraint vectors by graph neural networks and embedded into the attribute representations of the corresponding entities.
[0036] Hard constraint rules are expressed in predicate logic form, specifically including: Hard constraint H1: Matching constraint between ship tonnage and berth tonnage, expressed as... .
[0037] Hard constraint H2: Matching constraint between vessel length and shoreline length, for a set of vessels continuously berthed within the same shoreline. Satisfying the formula ,in For safety factor and ≥1.2, in this embodiment it is set to 1.2. The total length of the shoreline is 1533m. The first vessel occupies 216m of the shoreline, the second vessel occupies 144m, and the third vessel occupies 114m. The three vessels occupy a total of 474m, which is less than the total length of the shoreline of 1533m.
[0038] Hard constraint H3: Matching constraint between ship draft and effective water depth, denoted as... ,in For safety margin coefficient and =0.9, For a moment The tidal rise height; the tidal rise on that day was 1.2m, and the effective water depth was 15.1 + 1.2 = 16.3m. ×Effective water depth = 0.9 × 16.3 = 14.67m. The first vessel's draft is 10.5m, the second vessel's draft is 6.2m, and the third vessel's draft is 5.0m, all of which meet the requirements.
[0039] Hard constraint H4: Matching constraint between ship type and berth function, represented by berth function. Must belong to the ship type The required set of job qualifications; Hard constraint H5: Tidal window timing constraint, ship berthing time Mooring and departure times It must be located within the tidal window, among which For ships The moment of commencement of berthing, For ships The duration of operation; the tidal window is defined as a continuous period of time. ,in For ships The minimum safe tidal rise requirement. The tidal window for the day is from 10:00 to 15:00, and vessels must berth and depart within this window.
[0040] Soft constraint rules include: Soft constraint R1: Ship priority scheduling rules, based on the ship's operating type. Determine priority score The priority order is as follows: foreign trade liner vessels are the highest, followed by domestic trade liner vessels, then feeder barges, then foreign trade barges are the fourth, and domestic trade barges are the lowest. In this embodiment, the priority score for the first vessel is 3, the priority score for the second vessel is 2, and the priority score for the third vessel is 1.
[0041] Soft constraint R2: Operation time rationality rule, based on planned berthing time. Dock time with human experience The absolute value of the deviation (0.5 hours in this example) is used as the basis for the rationality score.
[0042] Soft constraint R3: Berth resource utilization maximization rule, based on berth space-time utilization rate. As the basis for scoring, the calculation formula is: ,in The total length of the berth shoreline. This represents the total duration of the statistical period.
[0043] Step S3: Based on the constraint rules in the hierarchical knowledge graph, perform feasibility screening on the ships to be scheduled and generate a feasible candidate set.
[0044] When performing feasibility screening based on constraint rules, a graph matching algorithm is used, combined with a constraint propagation mechanism, specifically including: Step S31: Represent the attributes of the vessel to be scheduled and the attributes of all candidate berths as attribute subgraphs in a hierarchical knowledge graph.
[0045] Step S32: Perform graph homomorphic matching to find berth nodes that match the ship attribute subgraph. The matching condition is that hard constraints H1, H2, H3, H4 and H5 are all met simultaneously.
[0046] The first vessel is a container ship, matched with a container berth. The tonnage requirement is 50,000 tons ≤ berth tonnage. All berth tonnage requirements are met, resulting in multiple candidate berths.
[0047] The second vessel is a bulk carrier, which is matched with bulk carrier berths, and multiple candidate berths are obtained.
[0048] The third vessel is a bulk carrier, which is matched with bulk carrier berths, and multiple candidate berths are obtained.
[0049] Step S33: During the matching process, constraint propagation is performed using the adjacency relationships of the knowledge graph: when the berth... When a berth is marked as occupied, the adjacent berths are automatically updated based on their adjacency relationships. and The available shoreline length is used to write the updated attributes back to the hierarchical knowledge graph.
[0050] Since no berths are currently occupied, there is no need to update the shoreline length of adjacent berths.
[0051] Step S34: Output all berth nodes that meet the matching conditions as a feasible berth candidate set. This set contains all berth indices that satisfy all hard constraints. .
[0052] Step S4: Based on the feasible candidate set, generate an evaluation value for each candidate solution to obtain the final evaluation value of each candidate solution.
[0053] Step S41: Generate multiple independent evaluation values for each candidate solution. Calculate priority scores respectively Time rationality score and resource utilization rate score Priority scoring Equal to the priority score of the ship Time rationality score ,in Resource utilization score (scale parameter set to 2) Equal to berth space utilization rate .
[0054] Step S42: Construct the feature vectors of candidate solutions. For candidate solutions... Its eigenvectors .
[0055] Step S43: Construct the state space of the reinforcement learning model. Includes the current port congestion index Number of ships to be dispatched The height of the water level during the next tidal window and equipment availability ; Step S44: Forward inference of the reinforcement learning model, concatenating the vectors. Input a reinforcement learning model, which employs a deep Q-network or policy gradient method, to maximize long-term cumulative reward. To optimize the objective, among which It is a discount factor with a value range of 0 to 1 (set to 0.95). The return is a single-step return and is defined as the increase in port throughput minus the penalty for ship delays. The model outputs the final evaluation value, which represents the total number of time steps. ; The training method for the reinforcement learning model is as follows: historical port scheduling data is used as training samples, and the throughput and delay time in the actual scheduling results are used as reward labels. The model parameters are obtained through offline training. In the online application stage, the model is updated incrementally using the scheduling data of the most recent week.
[0056] Step S45: Repeat step S44 for each candidate solution to obtain the final evaluation value of all candidate solutions.
[0057] Step S5: Determine whether the final evaluation value meets the requirements. If it does not meet the requirements, adjust the constraint rules and return to step S3. If it meets the requirements, proceed to step S6.
[0058] The specific method for determining whether the final evaluation value meets the requirements in step S5 is as follows: preset threshold. (0.7) Calculate the maximum value of the final evaluation of all candidate solutions. (0.82), where Indicates the index of all candidate solutions. Take the maximum value.
[0059] when When it is determined that the requirement is not met, The condition is then determined to meet the requirements; the specific method for adjusting the constraint rules is as follows: relax one non-critical hard constraint rule, which includes the safety factor of hard constraint H2. Safety margin factor of hard constraint H3 The relaxation method is to increase the safety factor. The value is updated to Safety margin factor The value is updated to ,in =0.05, =0.02. For example, the safety factor... The safety margin coefficient has been relaxed from 1.2 to 1.15. Relax the threshold from 0.9 to 0.88, then return to step S3 to re-filter.
[0060] Step S6: Select the optimal solution based on the final evaluation value of each candidate solution, and generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining the general large model.
[0061] The specific steps for generating natural language explanations using a general large model are as follows: Step S61: Retrieve hard constraint rule instances corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the conditions for satisfying each hard constraint, and generate the first explanatory text. The format of the text is: the ship name satisfies the tonnage requirements, draft requirements, length requirements, tidal window timing constraints and type matching requirements of the berth name. Step S62: Retrieve the soft constraint rules and the values of each evaluation value corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the final evaluation value output by the reinforcement learning model, and generate a second explanatory text, which includes priority score, time rationality score, resource utilization score and the final evaluation value output by the reinforcement learning model. Step S63: Combine the first and second explanation texts, and add the reasons for the final decision to generate a complete natural language explanation; Step S64: Output the natural language interpretation and scheduling scheme together to the human-computer interaction interface.
[0062] In this embodiment, based on the final evaluation values of each candidate scheme, the scheme with the highest evaluation value is selected: the first vessel is assigned to a container berth, the second vessel is assigned to a bulk cargo berth, and the third vessel is assigned to another bulk cargo berth. The planned berthing times are 10:00, 11:00, and 13:00, all within the tidal window of 10:00-15:00.
[0063] The general large model generates natural language interpretations following the steps described above: Execute step S61, retrieve the hard constraint rule instance corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the conditions for satisfying each hard constraint, and generate the first explanatory text: "The first vessel is a foreign trade container liner, which meets the tonnage requirements, draft requirements, length requirements and type matching requirements of the selected berth, and meets the berthing time requirements during the tidal window of 10:00-15:00." Execute step S62 to retrieve the soft constraint rules and evaluation values corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the final evaluation value output by the reinforcement learning model, and generate the second explanatory text: "Priority score 3.0, time rationality score 0.78, resource utilization score 0.85, after evaluation by the reinforcement learning model, the final score is 0.82, which is the optimal choice among all candidate schemes." In step S63, the first and second explanatory texts are concatenated, and the reasons for the final decision are added to generate a complete natural language explanation.
[0064] In step S64, the natural language explanation and scheduling plan are output to the human-computer interaction interface for the scheduler to confirm and execute.
[0065] Example 2 A system for constructing and integrating a port intelligent waterway knowledge graph includes: The knowledge graph construction module is used to build a basic knowledge graph for port and waterway transportation. It includes an entity extraction unit, an attribute binding unit, and a relationship construction unit. The entity extraction unit is used to extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. The attribute binding unit is used to bind key attributes to each entity. The relationship construction unit is used to establish semantic relationships between entities. The hierarchical fusion module is used to integrate port scheduling constraint rules onto the basic knowledge graph to form a hierarchical knowledge graph; the constraint rules include hard constraint rules and soft constraint rules. The feasibility screening module is used to screen the feasibility of ships to be scheduled based on the constraint rules in the hierarchical knowledge graph and generate a set of feasible candidates. The feasibility screening module adopts a graph matching algorithm and combines it with a constraint propagation mechanism. It uses the adjacency relationship of the knowledge graph to propagate constraints. When a berth is marked as occupied, it automatically updates the available shoreline length of adjacent berths. The evaluation value generation module is used to generate an evaluation value for each candidate solution based on the feasible candidate set, and obtain the final evaluation value of each candidate solution. The evaluation value generation module includes an independent evaluation value calculation unit, a feature vector construction unit, a state space construction unit, and a reinforcement learning inference unit. The judgment and adjustment module is used to determine whether the final evaluation value meets the requirements. If it does not meet the requirements, the constraint rules are adjusted and the feasibility screening module is triggered to regenerate the feasible candidate set. If it meets the requirements, the optimal solution selection module is triggered. The optimal solution selection module is used to select the optimal solution based on the final evaluation value of each candidate solution, and to generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining a general large model.
[0066] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.
[0067] Therefore, the present invention adopts the above-mentioned method and system for constructing and integrating a port smart waterway knowledge graph. Through the knowledge graph, the explicit expression of scheduling rules and the fusion of multi-source data are realized. Through reinforcement learning and feedback iteration mechanism, global optimization decision-making is realized. Through a general large model, an interpretable scheduling scheme is generated, which effectively improves the intelligence level and decision-making efficiency of port scheduling.
[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for constructing and integrating a port intelligent waterway knowledge graph, characterized in that, Includes the following steps: Step S1: Construct a basic knowledge graph of port and waterway transportation; Step S2: Integrate the port scheduling constraint rules into the basic knowledge graph to form a hierarchical knowledge graph; Step S3: Based on the constraint rules in the hierarchical knowledge graph, perform feasibility screening on the ships to be scheduled and generate a feasible candidate set; Step S4: Based on the feasible candidate set, generate an evaluation value for each candidate solution to obtain the final evaluation value of each candidate solution; Step S5: Determine whether the final evaluation value meets the requirements. If it does not meet the requirements, adjust the constraint rules and return to step S3. If it meets the requirements, proceed to step S6. Step S6: Select the optimal solution based on the final evaluation value of each candidate solution, and generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining the general large model.
2. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 1, characterized in that, Step S1 specifically includes the following steps: Step S11: Entity extraction. Extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. Ship entities include ship name and IMO number identification information; berth entities include berth name and port area identification information. Step S12: Attribute binding, binding key attributes to each entity; the attributes of the ship entity include the overall length of the ship. Maximum draft , load capacity tons and ship type ,in For ship indexing, indicating the first A vessel; attributes of the berth entity include berth length. Berth depth berth tonnage and berth function ,in For berth index, indicating the berth One berth; Step S13: Relationship construction, establishing semantic relationships between entities, including berthing relationships from ships to berths, transportation relationships from ships to cargo, and adjacency relationships from berths to adjacent berths.
3. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 1, characterized in that, The constraint rules in step S2 include hard constraint rules and soft constraint rules; the rules in the rule layer are encoded into constraint vectors by a graph neural network and embedded into the attribute representation of the corresponding entity.
4. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 3, characterized in that, Hard constraint rules are expressed in predicate logic form, specifically including: Hard constraint H1: Matching constraint between ship tonnage and berth tonnage, expressed as... ; Hard constraint H2: Matching constraint between vessel length and shoreline length, for a set of vessels continuously berthed within the same shoreline. Satisfying the formula ,in For safety reasons, The total length of the shoreline; Hard constraint H3: Matching constraint between ship draft and effective water depth, denoted as... ,in For safety margin coefficient, For a moment The height of the tidal rise; Hard constraint H4: Matching constraint between ship type and berth function, represented by berth function. Must belong to the ship type The required set of job qualifications; Hard constraint H5: Tidal window timing constraint, ship berthing time Mooring and departure times It must be located within the tidal window, among which For ships The moment of berthing begins For ships The duration of operation; the tidal window is defined as a continuous period of time. ,in For ships The minimum safe tidal rise requirement.
5. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 4, characterized in that, Soft constraint rules include: Soft constraint R1: Ship priority scheduling rules, based on the ship's operating type. Determine priority score The priority order is: foreign trade liner ships are the highest, followed by domestic trade liner ships, then feeder barges, then foreign trade barges are the fourth highest, and domestic trade barges are the lowest. Soft constraint R2: Operation time rationality rule, based on planned berthing time. Dock time with human experience The absolute value of the deviation is used as the basis for the reasonableness score; Soft constraint R3: Berth resource utilization maximization rule, based on berth space-time utilization rate. As the basis for scoring, the calculation formula is: ,in The total length of the berth shoreline. This represents the total duration of the statistical period.
6. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 5, characterized in that, In step S3, when performing feasibility screening based on constraint rules, a graph matching algorithm is used, combined with a constraint propagation mechanism, specifically including: Step S31: Represent the attributes of the vessel to be scheduled and the attributes of all candidate berths as attribute subgraphs in a hierarchical knowledge graph; Step S32: Perform graph homomorphic matching to find berth nodes that match the ship attribute subgraph. The matching condition is that hard constraints H1, H2, H3, H4 and H5 are all met simultaneously. Step S33: During the matching process, constraint propagation is performed using the adjacency relationships of the knowledge graph: when the berth... When a berth is marked as occupied, the adjacent berths are automatically updated based on their adjacency relationships. and The available shoreline length is used to write the updated attributes back to the hierarchical knowledge graph. Step S34: Output all berth nodes that meet the matching conditions as a feasible berth candidate set. This set contains all berth indices that satisfy all hard constraints. .
7. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 1, characterized in that, Step S4 specifically includes the following steps: Step S41: Generate multiple independent evaluation values for each candidate solution. Calculate priority scores respectively Time rationality score and resource utilization rate score Priority scoring Equal to the priority score of the ship Time rationality score ,in Resource utilization rate score is a scale parameter. Equal to berth space utilization rate ; Step S42: Construct the feature vectors of candidate solutions. For candidate solutions... Its eigenvectors ; Step S43: Construct the state space of the reinforcement learning model. Includes the current port congestion index Number of ships to be dispatched The height of the water level during the next tidal window and equipment availability ; Step S44: Forward inference of the reinforcement learning model, concatenating the vectors. Input a reinforcement learning model, which employs a deep Q-network or policy gradient method, to maximize long-term cumulative reward. To optimize the objective, among which As a discount factor, The return is a single-step return and is defined as the increase in port throughput minus the penalty for ship delays. The model outputs the final evaluation value, which represents the total number of time steps. ; Step S45: Repeat step S44 for each candidate solution to obtain the final evaluation value of all candidate solutions.
8. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 1, characterized in that, The specific method for determining whether the final evaluation value meets the requirements in step S5 is as follows: preset threshold. Calculate the maximum value of the final evaluation of all candidate solutions. ,in Indicates the index of all candidate solutions. Take the maximum value; when When it is determined that the requirement is not met, The condition is then determined to meet the requirements; the specific method for adjusting the constraint rules is as follows: relax one non-critical hard constraint rule, which includes the safety factor of hard constraint H2. Safety margin factor of hard constraint H3 The relaxation method is to increase the safety factor. The value is updated to Safety margin factor The value is updated to .
9. The method for constructing and integrating a port intelligent waterway knowledge graph according to claim 1, characterized in that, The specific steps for generating natural language explanations using the general large model in step S6 are as follows: Step S61: Retrieve hard constraint rule instances corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the conditions for satisfying each hard constraint, and generate the first explanatory text. The format of the text is: the ship name satisfies the tonnage requirements, draft requirements, length requirements, tidal window timing constraints and type matching requirements of the berth name. Step S62: Retrieve the soft constraint rules and the values of each evaluation value corresponding to the optimal scheduling scheme from the hierarchical knowledge graph, extract the final evaluation value output by the reinforcement learning model, and generate a second explanatory text, which includes priority score, time rationality score, resource utilization score and the final evaluation value output by the reinforcement learning model. Step S63: Combine the first and second explanation texts, and add the reasons for the final decision to generate a complete natural language explanation; Step S64: Output the natural language interpretation and scheduling scheme together to the human-computer interaction interface.
10. A system for constructing and integrating a port intelligent waterway knowledge graph, characterized in that, The method for constructing and integrating a port smart waterway knowledge graph as described in any one of claims 1-9 includes: The knowledge graph construction module is used to build a basic knowledge graph for port and waterway transportation. It includes an entity extraction unit, an attribute binding unit, and a relationship construction unit. The entity extraction unit is used to extract ship entities, berth entities, cargo entities, waterway entities, and weather event entities from multi-source heterogeneous data. The attribute binding unit is used to bind key attributes to each entity. The relationship construction unit is used to establish semantic relationships between entities. The hierarchical fusion module is used to integrate port scheduling constraint rules onto the basic knowledge graph to form a hierarchical knowledge graph; the constraint rules include hard constraint rules and soft constraint rules. The feasibility screening module is used to screen the feasibility of ships to be scheduled based on the constraint rules in the hierarchical knowledge graph and generate a set of feasible candidates. The feasibility screening module adopts a graph matching algorithm and combines it with a constraint propagation mechanism. It uses the adjacency relationship of the knowledge graph to propagate constraints. When a berth is marked as occupied, it automatically updates the available shoreline length of adjacent berths. The evaluation value generation module is used to generate an evaluation value for each candidate solution based on the feasible candidate set, and obtain the final evaluation value of each candidate solution. The evaluation value generation module includes an independent evaluation value calculation unit, a feature vector construction unit, a state space construction unit, and a reinforcement learning inference unit. The judgment and adjustment module is used to determine whether the final evaluation value meets the requirements. If it does not meet the requirements, the constraint rules are adjusted and the feasibility screening module is triggered to regenerate the feasible candidate set. If it meets the requirements, the optimal solution selection module is triggered. The optimal solution selection module is used to select the optimal solution based on the final evaluation value of each candidate solution, and to generate an optimal scheduling solution that includes berth allocation, time window and natural language interpretation by combining a general large model.