Port train scheduling method and device, computer equipment and readable storage medium

Abstract

The invention relates to a port train scheduling method and device, computer equipment and a readable storage medium. Comprising the following steps: acquiring real-time dynamic scheduling state information of the whole process of port train operation; performing graph embedding on the real-time dynamic scheduling state information to obtain a dynamic coupling topological graph of the whole process of port train operation; the dynamic coupling local topological graph is input into a space-time diagram convolutional network model, and a port train operation resource matching joint probability matrix is obtained through prediction; performing modeling by matching port train operation resources with the joint probability matrix to obtain a port train operation low-risk access dynamic feature set; carrying out safety constraint-oriented collaborative scheduling optimization on the port train operation low-risk access dynamic feature set to obtain a port train operation collaborative scheduling scheme; and selecting a target port train scheduling scheme which passes the efficiency simulation verification from the port train operation cooperative scheduling scheme. By adopting the method, the dispatching effect of the port train is improved.

Description

Technical Field

[0001] This application relates to the field of port train scheduling technology, and in particular to a port train scheduling method, apparatus, computer equipment, and computer-readable storage medium. Background Technology

[0002] With the continuous development of technology, port train scheduling technology is also constantly iterating. Currently, a general framework has been formed that combines prediction modules with heuristic or reinforcement learning scheduling and real-time cyclical distribution, realizing the updating of train and track plans at the second to minute level, thereby alleviating congestion and control problems to a certain extent.

[0003] Currently, port train scheduling systems mostly capture snapshots of the scene at fixed time steps and then independently complete scheduling in discrete optimization or reinforcement learning solvers. However, since train arrivals, loading and unloading progress, and track closure status can change abruptly in milliseconds, there is always a lag between the snapshot and the actual state. This means that once the solution is implemented, it may become out of touch with the actual conditions, which can easily trigger secondary ticket changes. At the same time, a large number of hypothetical conflicts and actual conflicts are mixed together, which can also lead to excessive shrinkage of the scheduling space. Therefore, the current scheduling of port trains is ineffective. Summary of the Invention

[0004] Therefore, it is necessary to provide a port train scheduling method, apparatus, computer equipment, and computer-readable storage medium to address the aforementioned technical problems and improve the efficiency of port train scheduling.

[0005] Firstly, this application provides a port train scheduling method, including:

[0006] Obtain real-time dynamic scheduling status information for the entire port train operation process;

[0007] The real-time dynamic scheduling status information is embedded into a graph to obtain a dynamic coupled topology graph of the entire port train operation process;

[0008] By performing threshold clipping on the dynamic coupled topology graph, a dynamic coupled local topology graph is obtained. Then, by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, the occupancy and delay probabilities of each topology node within a preset time window are predicted to obtain a joint probability matrix for port train operation resource matching.

[0009] By matching the port train operation resources with the joint probability matrix, the port train operation risk space modeling and safe path mining are performed to obtain the dynamic feature set of low-risk paths for port train operations.

[0010] A collaborative scheduling optimization based on safety constraints is performed on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations.

[0011] Select the target port train scheduling scheme that has passed the performance simulation verification from the aforementioned port train operation collaborative scheduling scheme.

[0012] Secondly, this application also provides a port train dispatching device, comprising:

[0013] The acquisition module is used to acquire real-time dynamic scheduling status information of the entire port train operation process;

[0014] The graph embedding module is used to embed the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph of the entire port train operation process.

[0015] The prediction module is used to obtain a dynamic coupled local topology graph by performing threshold clipping on the dynamic coupled topology graph, and to predict the occupancy and delay probabilities of each topology node within a preset time window by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, thereby obtaining a joint probability matrix for port train operation resource matching.

[0016] The mining module is used to perform risk space modeling and safe passage mining of port train operations by matching the port train operation resources with the joint probability matrix, and to obtain a dynamic feature set of low-risk passages for port train operations.

[0017] The scheduling optimization module is used to perform safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations.

[0018] The selection module is used to select a target port train scheduling scheme that has passed the performance simulation verification from the port train operation collaborative scheduling scheme.

[0019] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0020] The process involves: acquiring real-time dynamic scheduling status information for the entire port train operation process; embedding the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph for the entire port train operation process; performing threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph; inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model to predict the occupancy and delay probabilities of each topology node within a preset time window to obtain a joint probability matrix for port train operation resource matching; matching the port train operation resources with the joint probability matrix to perform port train operation risk space modeling and safety path mining to obtain a dynamic feature set of low-risk paths for port train operations; performing safety constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk paths for port train operations to obtain a collaborative scheduling scheme for port train operations; and selecting a target port train scheduling scheme that has passed performance simulation verification from the collaborative scheduling scheme for port train operations.

[0021] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0022] The process involves: acquiring real-time dynamic scheduling status information for the entire port train operation process; embedding the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph for the entire port train operation process; performing threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph; inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model to predict the occupancy and delay probabilities of each topology node within a preset time window to obtain a joint probability matrix for port train operation resource matching; matching the port train operation resources with the joint probability matrix to perform port train operation risk space modeling and safety path mining to obtain a dynamic feature set of low-risk paths for port train operations; performing safety constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk paths for port train operations to obtain a collaborative scheduling scheme for port train operations; and selecting a target port train scheduling scheme that has passed performance simulation verification from the collaborative scheduling scheme for port train operations.

[0023] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0024] The process involves: acquiring real-time dynamic scheduling status information for the entire port train operation process; embedding the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph for the entire port train operation process; performing threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph; inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model to predict the occupancy and delay probabilities of each topology node within a preset time window to obtain a joint probability matrix for port train operation resource matching; matching the port train operation resources with the joint probability matrix to perform port train operation risk space modeling and safety path mining to obtain a dynamic feature set of low-risk paths for port train operations; performing safety constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk paths for port train operations to obtain a collaborative scheduling scheme for port train operations; and selecting a target port train scheduling scheme that has passed performance simulation verification from the collaborative scheduling scheme for port train operations.

[0025] The aforementioned port train scheduling method, apparatus, computer equipment, and computer-readable storage medium first acquire real-time dynamic scheduling status information of the entire port train operation process; then, they perform graph embedding on the real-time dynamic scheduling status information to obtain a dynamic coupled topology graph of the entire port train operation process; next, they perform threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph; and by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, they predict the occupancy and delay probabilities of each topology node within a preset time window to obtain a joint probability matrix for port train operation resource matching; finally, they perform port train operation risk space modeling by matching the port train operation resources with the joint probability matrix. By mining safe pathways, a dynamic feature set of low-risk pathways for port train operations is obtained. Then, a collaborative scheduling optimization based on safety constraints is performed on this dynamic feature set to obtain a collaborative scheduling scheme for port train operations. Finally, a target port train scheduling scheme that has passed performance simulation verification is selected from the collaborative scheduling scheme. Based on this, in the dynamic and uncertain port train scheduling environment, the short-term risk prediction results and scheduling decision-making process can be linked in real time, directly outputting a target port train scheduling scheme with minimal conflicts and no need for frequent manual ticket changes. This allows for continuous matching of the on-site rhythm and stable throughput within a short timescale, thus improving the effectiveness of port train scheduling. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating a port train scheduling method in one embodiment;

[0028] Figure 2 This is a schematic diagram of the overall process of scheduling port trains in one embodiment of the port train scheduling method.

[0029] Figure 3 This is a schematic diagram comparing the application algorithm of the port train scheduling method in one embodiment with a traditional algorithm;

[0030] Figure 4 This is a structural block diagram of a port train dispatching device in one embodiment;

[0031] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0033] First, it should be understood that existing scheduling systems mostly capture snapshots of the situation at fixed time steps, and then independently complete scheduling in discrete optimization or reinforcement learning solvers. Since train arrivals, loading / unloading progress, and track closure status can change abruptly within milliseconds, there is always a lag between the snapshot and the actual state. Once the solution is implemented, it may become out of sync with the actual conditions, triggering secondary ticket changes. To buffer these discrepancies, the industry typically uses high penalty values ​​or multi-scenario Monte Carlo sampling to improve the robustness of the solution. However, these approaches increase computational load and mix a large number of hypothetical conflicts with actual conflicts, leading to an excessive contraction of the scheduling space. On the other hand, in terms of algorithms, the prediction module and the scheduling solver are often loosely coupled through interface files. Risk information is only reflected as a penalty term in the objective function and does not form rigid constraints during the action generation process. When track interlock chains or equipment failures spread rapidly, heuristic search and deep reinforcement learning strategies need to repeatedly try and fail to perceive new feasible boundaries, resulting in an output speed that cannot match the actual pace of the situation. Meanwhile, to avoid safety incidents, reinforcement learning models have to set extremely high conflict penalties, resulting in a conservative waiting strategy and a lower throughput rate. Overall, existing methods still follow a two-stage structure of probabilistic prediction and penalty-based optimization, lacking an algorithmic mechanism to directly map short-term risks into operable scheduling boundaries. Therefore, there is an urgent need for a port train scheduling method that can improve the effectiveness of port train scheduling.

[0034] In one embodiment, such as Figure 1As shown, a port train scheduling method is provided. This embodiment uses the method applied to a terminal as an example. The terminal includes, but is not limited to, personal computers, laptops, smartphones, and tablets. The terminal includes an acquisition module, a graph embedding module, a prediction module, a mining module, a scheduling optimization module, and a selection module. The acquisition module acquires real-time dynamic scheduling status information of the entire port train operation process. The graph embedding module embeds the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph of the entire port train operation process. The prediction module performs threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph. It also inputs the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model to predict the occupancy and delay probabilities of each topology node within a preset time window, obtaining a joint probability matrix for port train operation resource matching. The mining module... The module is used to model the operational risk space and mine safe paths of port trains by matching port train operation resources with a joint probability matrix, thereby obtaining a dynamic feature set of low-risk paths for port train operations. The scheduling optimization module is used to perform safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk paths for port train operations, thereby obtaining a collaborative scheduling scheme for port train operations. The selection module is used to select a target port train scheduling scheme that has passed the performance simulation verification from the collaborative scheduling scheme for port train operations. In this way, in the dynamic and uncertain port train scheduling environment, the short-term risk prediction results and the scheduling decision process can be linked in real time, directly outputting a target port train scheduling scheme with the fewest conflicts and without frequent manual ticket changes. This allows for continuous matching of the on-site rhythm and maintaining stable throughput within a short time scale, thus achieving the goal of improving the scheduling effect of port trains. In this embodiment, the method includes the following steps 202 to 212:

[0035] Step 202: Obtain real-time dynamic scheduling status information for the entire port train operation process.

[0036] It should be noted that the real-time dynamic scheduling status information may include three types of asynchronous information: dock loading and unloading plan, train operation instructions, and loading and unloading equipment status. In other words, each status change is written to a millisecond-level event queue, and the rolling status tables of trains, tracks, and equipment are refreshed synchronously.

[0037] This includes obtaining real-time dynamic scheduling status information for the entire port train operation process, including:

[0038] The system acquires real-time status information from asynchronous inputs throughout the entire port train operation process, and synchronizes this real-time status information to a millisecond-level event sequence using a unified event identifier, resulting in a millisecond-level ordered event sequence. This ordered event sequence is then expanded according to the subject dimension and written into a preset event rolling status table to obtain the rolling status information of the port train operation object and the constraint factors affecting the port train operation. Based on the rolling status information and constraint factors, a real-time status vector for the entire port train operation process is generated. Finally, the real-time status information is mapped to the scheduling base information flow to obtain real-time dynamic scheduling status information.

[0039] Specifically, the steps for obtaining real-time status information of the asynchronous input of the entire port train operation process, and synchronizing the real-time status information to a millisecond-level event sequence according to a unified event identifier to obtain a millisecond-level ordered event sequence can be as follows:

[0040] The system simultaneously integrates the terminal's loading and unloading bridge operation plans, container destinations, and shore-side operation schedules with train arrival forecasts, route instructions, and track closure announcements generated by the centralized train dispatching system. It also receives equipment location, load, and fault alarms from the gantry crane monitoring system in parallel. A unified event identifier is used. All changes are written sequentially to a millisecond-level event queue. ,in It is an event code that uniquely points to a single change in the scene; An ordered queue for storing a sequence of events, the internal order of which is the order in which changes occur.

[0041] It is understandable that by uniformly processing the asynchronous inputs of the three types of source systems, it is ensured that trains, tracks, and equipment are captured at the same moment when a sudden change in state occurs, thus laying a continuous and gapless dynamic information foundation for conflict identification.

[0042] The steps involved in expanding the millisecond-level ordered event sequence by subject dimension and writing it into a preset event rolling state table to obtain the rolling state information of the port train operation object and the constraint factor information affecting the port train operation, and generating the real-time state vector of the entire port train operation process based on the rolling state information and constraint factor information, can be as follows:

[0043] For event queues Expand and write to the scrolling status table according to the subject dimension. ; lower the main body of the train to the train carriage The main body of the stock market will be placed in the stock market. The main body of the equipment is placed on the equipment row. And attach a unique timestamp to each record. and business tags; also in the appendix The continuous recording of the duration of blockades, rainfall, and faults allows any constraints to be read instantly during the scheduling phase, forming a highly timely state vector for conflict calculation. For the set of train numbers; For the set of stock track numbers; A set of numbers for loading and unloading equipment; The moment the event occurred; The main table shows the latest status of the main entity; Appendix table for critical restriction intervals.

[0044] The steps for mapping real-time status information to the scheduling base information stream to obtain real-time dynamic scheduling status information can be as follows:

[0045] The state vector is mapped and written into the scheduling base information stream according to predetermined fields. Maintain consistency between fields, enumerations, and encodings; among them, The structured information flow is shared by all downstream modules, and its content is... main table and The attached table shows the real-time composite results after mapping the unified fields.

[0046] Step 204: Graph embedding is performed on the real-time dynamic scheduling status information to obtain a dynamic coupled topology graph of the entire port train operation process.

[0047] It should be noted that, based on the status table, a physical-time dual-layer node topology can be constructed within a unified field reference system. The track and yard operation area can be flexibly segmented into variable lengths, loading and unloading coupling nodes and fault gate control nodes can be set to generate an integrated topology map with dynamic probability weights. The integrated topology map with dynamic probability weights is the dynamic coupling topology map.

[0048] Among them, graph embedding is performed on the real-time dynamic scheduling status information to obtain a dynamic coupled topology graph of the entire port train operation process, including:

[0049] Physical nodes are allocated to the job resource objects corresponding to the real-time dynamic scheduling status information, and each physical node is time-replicated according to a preset time window to obtain a two-layer node mapping feature with physical and temporal attributes. Based on the two-layer node mapping feature, the job resource objects are partitioned according to a flexible segmentation strategy to obtain the flexible nodes corresponding to the port trains, and the set of occupied flexible segments of the flexible nodes is matched in a coverage minimization manner. The corresponding blocking weights are matched to the job dynamic constraint information in the set of occupied flexible segments to obtain the structured topology constraint information of the port trains. According to the structured topology constraint information, the reachability combination of all job nodes in the set of occupied flexible segments is updated to obtain the set of job node edges. By performing short-term event statistics on the set of job node edges, the probability vector of job node edges of the port trains is obtained. According to the probability vector of job node edges, a dynamic coupled topology graph is generated.

[0050] Specifically, the steps of allocating physical nodes to job resource objects corresponding to real-time dynamic scheduling status information, and performing time replication on each physical node according to a preset time window to obtain a two-layer node mapping feature with physical and time attributes can be as follows:

[0051] The output coordinates of train, track, junction, storage yard, and loading / unloading equipment are uniformly converted to the on-site reference system. Assign a unique physical node to each resource entity. Uses a continuously scrolling five-minute time window. Perform time replication on each physical node, at time... Generation time node , where window number Calculated by the following formula:

[0052]

[0053] in, The current moment; This serves as the reference time for the rolling window; The window width is five minutes; based on the above formula, integer division ensures that any real-time moment is uniquely mapped to the corresponding time node, thus forming the physical layer. -Time layer A two-tier node architecture; then, an index mapping is written to each time node. .

[0054] Understandably, the above method enables scheduling reasoning to both trace back the location of entities and pinpoint specific time periods, ensuring that conflict capture is aligned with the on-site rhythm at a short time granularity.

[0055] The steps of using the two-layer node mapping features to partition the operational resource object according to the flexible segmentation strategy to obtain the flexible nodes corresponding to the port train, and matching the set of occupied flexible segments of the flexible nodes in a coverage minimization manner can be as follows:

[0056] For track nodes in a two-layer node framework The variable occupancy length strategy is used to divide the track into sections based on the actual length of the track. Then, based on the minimum segmentable length defined by the system... and train formation length set Calculate the number of flexible segments This allows for the construction of flexible nodes. ;

[0057] Flexible nodes are generated using the same method for both the yard operation area and the loading / unloading equipment boom, allowing for a unified measurement of heterogeneous resource occupancy. Further, for each train... Calculate its occupied segment set The following formula is given

[0058]

[0059] in, For train The minimum flexible segment coverage set; For flexible segment The physical length; For train Group length.

[0060] Understandably, the above steps can match train lengths and flexible segment sets in a way that minimizes coverage, ensuring that both long and short train sets can be accurately occupied within the same topological framework, thus eliminating scheduling misalignments caused by length differences.

[0061] The steps for matching the dynamic constraint information of the operation in the set of flexible sections with the corresponding blocking weights to obtain the structured topology constraint information of the port train can be as follows:

[0062] Cross-domain restrictions between loading / unloading equipment and vehicle resources are implemented using virtual coupling nodes. The method of embedding a double-layer structure first involves each pair of yard cranes Adjacent tracks Generate loading and unloading coupling nodes And establish two directed edges. , This is used to express the interlocking relationship where crane operation blocks track passage; subsequently, each potentially malfunctioning piece of equipment... Generate fault gate nodes Instruct it to read the device status flags. ,when Automatically close the connection in case of (fault). Simultaneously improve all with Blocking weights of adjacent edges.

[0063] Understandably, through the above steps of coupling-gating design, loading and unloading constraints, equipment failures, and track blockages are natively written into the topology, and scheduling search can avoid potential conflicts without additional judgment, thus achieving structured absorption of dynamic constraints.

[0064] The steps for updating the reachability combination of all job nodes occupying the flexible segment set based on the structured topology constraint information to obtain the job node edge set can be as follows:

[0065] Enumerate the possible combinations for the complete set of nodes Nph∪Ntm∪Nfx∪Ncp in sequence.<u,v> Combinations of physical adjacency, temporal continuity, or coupled relationships are constructed into directed edges eu→v∈

[0066] For each edge, a passable time window ωu→v and a job conflict coefficient κu→v are written, and these two attributes are overwritten and updated by the event-driven function Ψ when the real-time event set Θ(t) is triggered. The core rule is given by the following formula:

[0067] [ωu→v(t), κu→v(t)] = Ψ(ωu→v(t-1), κu→v(t-1), Θ(t))

[0068] Where ωu→v(t) is the latest passable time window of edge eu→v at time t; κu→v(t) is the latest conflict coefficient of edge eu→v at time t; and Θ(t) is the set of on-site events occurring at time t (including train delays, lifting of blockades, sudden rainfall, etc.). This is an instant mapping function that reduces the number of passes based on the event type and accumulates the conflict coefficient.

[0069] Understandably, by directly rewriting ω and κ at the moment an event occurs, the two types of business impacts—shortening the remaining window due to arrival delay and raising the conflict threshold due to equipment failure—are embedded into the edge attributes, ensuring that the scheduling engine can instantly perceive the risk boundary when searching for paths, without needing to additionally judge multi-source restrictions.

[0070] The step of obtaining the port train's operation node edge probability vector by performing short-term event statistics on the operation node edge set can be as follows:

[0071] In the dynamically updated edge set The top sliding window for the last fifteen minutes The system counts occupancy, delay, and interlock events, and maps the results to a ternary probability vector. :

[0072] p e (t)= 1 | W | ∑ τ∈ W [ I oc (e,τ), I dl (e,τ), I il (e,τ)]

[0073] in, For a moment side Trivariate probability vector [ p e oc ,  p e dl ,  p e il ] ; The number of discrete time points within the window; , , To indicate the edges respectively At any moment Boolean functions for determining whether a train is in use, whether it is delayed due to arrival, or whether it is in use due to interlocking.

[0074] Understandably, by using the above steps to characterize risk intensity with window frequency rather than statistical mean, a significant probability difference is created between normal stock market congestion and occasional blockage; ultimately... By writing edge attributes, dynamic weighting of the integrated topological probability graph of tracks, storage yards, equipment, and time is completed, providing a structured risk quantification basis for the next step of short-term forecasting.

[0075] The steps for generating a dynamically coupled topology graph based on the edge probability vectors of the task nodes can be as follows:

[0076] According to the node set Edge set With probability weights The three items should be output in a unified format file. A fixed field order and coding standard are adopted to ensure the connection with step S3, and the source and function of physical nodes, flexible nodes and coupling nodes are marked in the metadata to facilitate the rapid loading of downstream models.

[0077] The above steps encapsulate the three types of business pain points—train formation differences, loading and unloading interlocking, and fault blocking and contraction—into graphical data in a structured weighted manner, ensuring that subsequent short-term forecasting and scheduling strategies directly address real conflicts.

[0078] Step 206: By performing threshold clipping on the dynamic coupled topology graph, a dynamic coupled local topology graph is obtained. Then, by inputting the dynamic coupled local topology graph into the spatiotemporal graph convolutional network model, the occupancy and delay probabilities of each topology node within a preset time window are predicted to obtain the joint probability matrix for port train operation resource matching.

[0079] It should be noted that threshold clipping can be performed on the topology graph to obtain local subgraphs. A spatiotemporal graph convolutional network is used to predict the occupancy probability and delay probability of each node within a preset time window in the future, and the train-track joint probability matrix is ​​output. That is, the joint probability matrix is ​​obtained by outputting the train-track joint probability matrix.

[0080] Specifically, by performing threshold edge pruning on the dynamically coupled topology graph, a dynamically coupled local topology graph is obtained. Furthermore, by inputting this dynamically coupled local topology graph into a spatiotemporal graph convolutional network model, the occupancy and delay probabilities of each topology node within a preset time window are predicted, resulting in a joint probability matrix for port train operation resource matching, including:

[0081] The dynamic coupling topology graph is input into a preset edge threshold layer to obtain the scalar values ​​of the workload and conflict weight of each directed edge in the dynamic coupling topology graph. All workload and conflict weight scalar values ​​are identified by a preset dynamic threshold function, and the available directed edges of all directed edges in the dynamic coupling topology graph are identified. By merging all available directed edges in the same direction, a dynamic coupling local topology graph is obtained.

[0082] Specifically, the steps for inputting the dynamic coupling topology graph into a preset edge threshold layer to obtain the scalar values ​​of the workload and conflict weight for each directed edge in the dynamic coupling topology graph can be as follows:

[0083] Dynamically Coupled Topology Diagram Input adaptive edge thresholding layer, for each directed edge Read the current job load scalar Conflict weight scalar ;

[0084] Using dynamic threshold function right Perform online judgment and remove all those below [the specified value]. And the side will not affect the safe passage of trains in the next 10-30 minutes;

[0085] Perform a unidirectional merge on the retained edges, merging multiple edges of the same resource pair within a continuous time window into a single index, ultimately generating a local subgraph. .

[0086] Understandably, by using the two-step operation of threshold edge trimming and unidirectional merging, irrelevant nodes and redundant paths are directly compressed, so that subsequent inference focuses only on the three most conflict-sensitive structures: train entry and exit, loading and unloading connection, and blockade release, ensuring that the prediction target corresponds to the scheduling pain point.

[0087] Specifically, by using a preset dynamic threshold function to identify all job load scalar values ​​and conflict weight scalar values, the available directed edges in the dynamically coupled topology graph are identified, including:

[0088] For local subgraphs A spatiotemporal graph convolutional network is employed, concatenating the occupancy state vector, interlocking restriction code, and time and location information of each node into the input signal. After unfolding the adjacency matrix along the time dimension, multiple convolutions are performed to progressively propagate train propagation rhythm, track release rhythm, and cross-regional equipment interference within the network. After convolution, the outputs of each layer are normalized, mapping the occupancy probability and delay probability of each node within the next 10-30 minutes to continuous values ​​in the interval [0,1]. This approach processes spatial relationships coupled by a graph structure and captures temporal chain effects through convolution stacking, enabling the prediction results to directly reveal which track is most likely to be occupied or blocked by which train at which time, providing a quantitative basis for refined scheduling.

[0089] Among them, by merging all available directed edges in the same direction, a dynamically coupled local topology graph is obtained, including:

[0090] The convolutional output is matched along two axes according to the train number and the track number. A Cartesian mapping is then performed on the node occupancy probability within the same time slice to form a joint probability matrix of train arrival and track occupancy. ;right Elements exceeding the warning threshold are marked with a conflict flag and written to the interface file in the order of train number, track number, start and end time slices, and conflict intensity fixed fields. The matrix and its labels are directly read by the downstream risk grid to bypass high-risk combinations during scheduling searches, thereby providing train and track scheduling schemes that do not require large-scale ticket changes.

[0091] Step 208: By matching port train operation resources with a joint probability matrix, the port train operation risk space is modeled and the safe access path is mined to obtain the dynamic feature set of low-risk access paths for port train operations.

[0092] Specifically, by matching port train operation resources with a joint probability matrix, the operational risk space of port trains is modeled and safe access routes are mined, resulting in a dynamic feature set of low-risk access routes for port train operations, including:

[0093] The joint probability matrix is ​​decomposed line by line according to the predetermined route of the port train to obtain the operational constraint probability of the port train under each route. Based on the operational constraint probability and the corresponding on-site load weight, the comprehensive operational risk value of the port train under each route is generated. The path containing the comprehensive operational risk value is subjected to 3D raster mapping and interpolation completion processing to obtain the 3D continuous risk field corresponding to the port train. The 3D continuous risk field is subjected to risk region hierarchical fusion processing to obtain the high-risk block set of the port train. Risk block constraint extension processing is performed in the high-risk block set to obtain the safety buffer zone under the hard constraint of the operational resource object. The passable grid cells of the port train are selected from the safety buffer zone, and the risk perception scheduling boundary of the port train is extracted based on the shortest path search of the passable grid cells via the time layer. Based on the risk perception boundary, the remaining available time, remaining capacity and risk gradient information of the low-risk path are standardized to obtain the dynamic feature set of the low-risk path of the port train operation.

[0094] Specifically, the joint probability matrix is ​​decomposed line by line according to the predetermined routes of the port train to obtain the operational constraint probability of the port train under each route. Based on the operational constraint probability and the corresponding on-site load weight, the comprehensive operational risk value of the port train under each route is generated, including:

[0095] The joint probability matrix is ​​decomposed line by line according to the predetermined train route. For the first... Read the occupancy probability of each path. Arrival delay probability Probability of overlap with assignment Then, the three probabilities are multiplied by the on-site load weight. The overall risk value is calculated using a maximum-dominant strategy. :

[0096]

[0097] in, For the way The probability that the time period will be occupied by a train; For the way The probability of being blocked due to arrival delay; For the way The probability of overlapping with loading and unloading operations; The on-site load weights for the three types of risks are adjusted online according to shift density; For the way The overall risk value.

[0098] The above steps automatically amplify the effects of delays or overlaps during peak periods, prioritizing the paths that truly threaten throughput rhythm in subsequent spatial mapping; simultaneously... Together with the associated track number Crane number and predicted time slices Write the path record.

[0099] The process involves three-dimensional rasterization mapping and interpolation completion of the path containing the comprehensive operational risk value to obtain a three-dimensional continuous risk field corresponding to the port train, including:

[0100] For path records containing comprehensive risk values, spatial rasterization is performed by first mapping the coordinates of the route start and end nodes to a planar index. Then predict the time slice Mapped to vertical index Constructing a 3D mesh ; for each path Write the corresponding cell Linear interpolation is used to assign values ​​to lattice cells not directly covered by the route, generating a continuous risk field covering the three-dimensional domains of nodes, edges, and time. .

[0101] Understandably, the mapping operations described above transform discrete route risks into a visual grid, enabling morphological operations to handle cascading threats from multiple routes based on geometric adjacency relationships.

[0102] The steps involved in performing risk region hierarchical fusion processing on the three-dimensional continuous risk field to obtain a set of high-risk blocks for port trains include:

[0103] For continuous risk fields The dual-threshold classification method identifies risk values ​​exceeding the warning threshold. The cell is marked as high risk, and will be between With tolerance threshold Cells between these ranges are marked as bufferable, and the rest are marked as low-risk; further, the high-risk cell set is then... Perform morphological dilation to merge adjacent high-risk regions:

[0104]

[0105] in, This is the original set of high-risk lattice cells; It is a cubic structural element, and its side length is determined by the minimum safe distance on site; High-risk grid coordinates; The structuring element displacement vector; This is the set of high-risk blocks after expansion.

[0106] Understandably, through the analysis of With structural elements By performing Minkowski summation, spatially adjacent and temporally continuous high-risk cells are merged into a single hazard block, generating avoidance zones that are more in line with operator perception, thus establishing a unified geometric boundary for equipment window overlay and safety corridor extraction.

[0107] Specifically, risk block constraint extension processing is performed on the high-risk block set to obtain a safety buffer zone under the hard constraints of the job resource object, including:

[0108] The high-risk block set after morphological expansion Renumbered For each piece Call the device mapping function Search its covered resource set, including cranes, storage yards, and inspection stations. and read the corresponding shutdown window family. Ξ d ={[ a d,k , b d,k ]} .

[0109] Furthermore, a strategy for outward convex extension is proposed. The time axis is symmetrically expanded at both ends, and the expansion length is:

[0110] t s ' ( B i ) = t s ( B i )- max d ∈Ψ ( B i ) max [a,b] ∈ Ξ d b< t s ( B i ) ( t s ( B i )-b) t e ' ( B i ) = t e ( B i )+ max d ∈Ψ ( B i ) max [a,b] ∈ Ξ d a> t e ( B i ) (a- t e ( B i ))

[0111] in, Risk block The original start and end times; The start and end times after the expansion; To and The associated set of devices; For equipment A set of shutdown windows; [a,b] The start and end times of a single shutdown window.

[0112] Based on the above formula Find the end point of the most recent equipment shutdown on the left. Find the starting point for the next shutdown on the right. The maximum time difference on both sides is incorporated into the risk block, thus obtaining a safety buffer zone with hard constraints on equipment. B safe ={[ t s ' ( B i ), t e ' ( B i )]} .

[0113] Understandably, by following the steps described above, the scheduling strategy can ensure that trains or loading / unloading tasks are not pushed into blind spots where equipment is down or blocked, directly addressing the business pain point of avoiding on-site conflicts.

[0114] The steps for selecting passable grid cells for port trains from the safety buffer zone, and then extracting the risk-aware scheduling boundary of port trains based on the shortest path search with time layer through the passable grid cells, can be as follows:

[0115] Safety buffer zone outer perimeter to global grid Perform the distance transformation, first for each cell Calculate its Euclidean distance to the nearest safe buffer cell. Then, using the minimum safe distance threshold Select a passable set :

[0116]

[0117] in, For grid element arrive The shortest Euclidean distance; The minimum safe distance threshold specified on site; The set of passable lattice cells that meet the safety distance requirement.

[0118] exist The system employs a shortest path search with a time layer to connect all entry and exit nodes, extracting hierarchical schedulable corridors. At the same time, the risk gradient on both sides of the corridor will be adjusted. Write edge attributes to form risk-aware scheduling boundaries .

[0119] Understandably, by using distance filtering and shortest connectivity as two steps, the strategy learning phase can have enough room to maneuver while meeting the bottom line of safety, avoiding an imbalance in throughput due to excessive conservatism, and also preventing the entry into high-risk areas that could lead to secondary ticket changes.

[0120] The steps for standardizing the remaining available time, remaining capacity, and risk gradient information of low-risk routes based on the risk perception boundary to obtain the dynamic feature set of low-risk routes for port train operations can be as follows:

[0121] Risk perception scheduling boundary Including the remaining available time for each low-risk route With remaining capacity Write to standard interface file The fields are in a fixed order: path number, start and end time period, remaining track occupancy, remaining equipment load, and risk gradient.

[0122] Understandably, the output directly serves as an explicit hard constraint on the initial action space and reward function, preventing the reinforcement learning policy from exploring high-risk areas, thereby instantly transforming the prediction results into a safe space that is available to the scheduling side and dynamically shrinks or expands over time.

[0123] Step 210: Perform safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations.

[0124] It should be noted that the action space is limited by a low-risk corridor, and a train-track-time slice scheduling scheme is generated by reinforcement learning according to a two-level diffusion strategy of railway layer and loading and unloading layer. During the diffusion process, risk temperature scaling noise and scene memory similarity adjustment are used to achieve fast convergence, and gradient-driven local repair is performed on atomic actions that fall into high-risk cells.

[0125] Among them, a collaborative scheduling optimization based on safety constraints is performed on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations, including:

[0126] A two-layer action set between the railway layer and the loading / unloading layer is generated based on the dynamic feature set of the risk path. A traceable scheduling skeleton is constructed based on the inter-layer mapping relationship established by the two-layer action set. Based on the traceable scheduling skeleton, priority-oriented initial diffusion noise is assigned to the railway layer actions, resulting in the convergent train sequence for the railway layer and prior constraints for the loading / unloading layer actions. Under the convergent train sequence for the railway layer and the prior constraints for the loading / unloading layer actions, the three-dimensional comprehensive risk field is mapped to a temperature tensor. The amplitude of the action diffusion noise is dynamically adjusted through a temperature scaling model, resulting in a reinforcement learning safety exploration rule embedded in the safety boundary. Then, based on the three types of features of the diffusion mid-stage scene extracted by the reinforcement learning safety exploration rules, the scene fingerprint is spliced ​​to obtain the scene fingerprint. After matching the scene fingerprint with the historical memory database, the diffusion noise center of this round is updated according to the segment replacement rule. After the diffusion converges in the railway layer and the loading and unloading layer, the risk field back lookup is performed on the output port train operation collaborative scheduling scheme to obtain the risk field back lookup results. The risk field back lookup results include at least one candidate port train operation collaborative scheduling scheme. Among the candidate port train operation collaborative scheduling schemes, the port train operation collaborative scheduling scheme is selected according to business preferences.

[0127] Specifically, the steps for generating a two-layer action set between the railway layer and the loading / unloading layer based on the dynamic feature set of the risk path, and constructing a traceable scheduling skeleton based on the inter-layer mapping relationship established by the two-layer action set, can be as follows:

[0128] The aforementioned low-risk corridor By train number Stock Number Time slice Decompose and generate railway layer actions using set analysis:

[0129]

[0130] in, For railway layer action set; For a single railway action, by the train , stock channel Time slice Composed of triplets; For the first Train number; For the first Stock track number; For the first Time slice; This is a low-risk corridor cluster.

[0131] Then each one Retrieve the available crane window for the same time slot from the yard resource mapping table. Combined into loading and unloading operations And merge into the collection A two-level index table is used. Establish a one-to-many mapping between railway operations and all loading and unloading extensions to provide a scheduling framework that covers all time periods and is traceable between layers for diffusion.

[0132] The steps for obtaining the convergent train sequence at the railway level and the prior constraints on loading and unloading level actions, based on a traceable scheduling skeleton and prioritizing the initial diffused noise for railway level actions, can be as follows:

[0133] Actions on the railway level Initial diffusion noise is allocated, and a symmetrical distribution is assembled using a Gaussian kernel with train priority as the centroid. During the diffusion iteration, the convergent train sequence is recorded for each fluctuation sequence. and immediately write back to the index table. As a priori for loading and unloading layers; for all Independent Gaussian noise is applied to the parameter channels that are not locked by the railway layer, and the loading and unloading layer diffusion is activated. This makes the coarse-grained train passage and the fine-grained crane connection exhibit a layered characteristic of sequential absorption and sequential convergence, thus avoiding the simultaneous divergence of the two layers and causing global scheduling instability.

[0134] Among them, under the convergence of train sequences at the railway level and the prior constraints of actions at the loading and unloading level, the three-dimensional comprehensive risk field is mapped into a temperature tensor, and the amplitude of action diffusion noise is dynamically adjusted through a temperature scaling model to obtain the reinforcement learning safety exploration rules embedded in the safety boundary.

[0135] The aforementioned generated three-dimensional comprehensive risk field Mapped to temperature tensor Read any atomic action scalar temperature of the grid T(a) ∈ [0,1] ; Actions in the diffusion loop noise variance Temperature scaling model

[0136]

[0137] in, For action The noise variance of the current iteration step; The minimum noise limit set for the system is used for high-risk suppression; The maximum noise limit set for the system is used for low-risk relaxation; For action The temperature coefficient of the corresponding lattice cell is derived from .

[0138] Based on the above formula, it is possible to make the noise amplitude inversely proportional to the risk when... When approaching 0 (high-risk zone), Approaching This causes the diffusion path to be rerouted; when When approaching zone 1 (low-risk zone), near This encourages thorough searching; the buffer temperature can be kept at the median to allow for trial-and-error fine-tuning.

[0139] Understandably, the above steps can be based on the concept of risk temperature control, directly embedding the safety boundary into the noise generation mechanism to ensure that reinforcement learning exploration is neither overly conservative nor touches high-risk cells, thereby serving the business pain points of short-time train and track scheduling.

[0140] The steps for obtaining a scene fingerprint by concatenating three types of features from the diffusion mid-stage scene extracted based on reinforcement learning security exploration rules, and then updating the current round of diffusion noise center according to the segmented replacement rules after matching the scene fingerprint with the historical memory database can be as follows:

[0141] The current scene, which has diffused into the middle section, will be extracted into three types of features: train arrival flow, track closure mode, and equipment status group, which will be encoded into vectors respectively. , , Then, according to predetermined weights, they are spliced ​​together to form a scene fingerprint. f=[ f arr , f blk , f dev ] Then traverse the memory bank. Calculate the similarity for each sample. Select the maximum value and its corresponding solution center ;

[0142] A segmented replacement rule is proposed to update the noise center in this round. :

[0143]

[0144] in, Fingerprint for the current scene; For historical sample memory bank; For the current scene and memory samples Similarity; Enable thresholds for samples; The scheme center is the sample with the highest similarity. The default random center; This is the center of the diffused noise in this round.

[0145] By adopting the above steps, the strategy can only borrow mature solutions when the scenarios are highly similar, ensuring a balance between quickly approaching feasible solutions and maintaining innovative search, and fundamentally reducing the large-scale reordering triggered by sudden events.

[0146] The process of performing a risk field back-check on the output port train operation collaborative scheduling schemes after diffusion convergence at the railway and loading / unloading layers, and obtaining the risk field back-check results, wherein the risk field back-check results include at least one candidate port train operation collaborative scheduling scheme, can be described as follows:

[0147] After the double diffusion convergence of the railway layer and the loading / unloading layer, a risk field backcheck is performed on the output scheme; trains that still fall into high-risk cells are indexed into a set. All actions of trains outside the assembly were frozen; subsequently, actions were taken on each target train. Read the risk gradient of its cell and according to segmented step size Initiate localized scrolling diffusion, adjusting the entry and exit order only within a limited time window. This application proposes an iterative method for step size control based on the risk gradient:

[0148]

[0149] in, For train Risk gradient of the cell; High and low gradient thresholds;

[0150] This corresponds to the diffusion step size; For train The local diffusion step size.

[0151] Based on the above steps, by mapping the gradient magnitude to the step size, the patch correction is focused on the steep risk areas, which can quickly pull away dangerous blocks and avoid chain changes to the already stable global schedule, ensuring that the final solution eliminates residual conflicts without adding new tickets.

[0152] The steps for selecting the optimal port train operation collaborative scheduling scheme from among the candidate schemes based on business preferences can be as follows:

[0153] The multiple train L, track G, and time slice τ schemes, after flexible repairs, were compiled into a scheme cluster; a risk temperature spectrum was added to each scheme. Similarity to history The system tags the tickets and calculates the corresponding number of ticket changes and the remaining buffer time. The system uses fixed fields to write the data into a standard interface file, which is then used by step S6 to select the best option based on business preferences (such as prioritizing the fewest ticket changes or the most reliable option). This ensures that the scheduling instructions are both within the safety boundaries and take into account the cost of changing tickets on-site.

[0154] Step 212: Select the target port train scheduling scheme that has passed the performance simulation verification from the port train operation collaborative scheduling scheme.

[0155] It should be noted that the scheduling scheme is verified by parallel simulation. When it meets the throughput and safety thresholds, it is encapsulated into a scheduling instruction packet and sent to the dispatch console, locomotive and yard terminal. After the instruction is executed, the field deviation is continuously monitored. When the deviation exceeds the threshold, it triggers a backflow to step S3 for recalculation.

[0156] Among them, the target port train scheduling schemes that have passed the performance simulation verification are selected from the port train operation collaborative scheduling schemes, including:

[0157] The port train operation collaborative scheduling schemes are simulated one by one according to priority. Multiple models are called for parallel simulation and data is recorded. After automatic verification by dual baselines, at least one qualified port train operation collaborative scheduling scheme is selected. The qualified port train operation collaborative scheduling schemes are simulated and executed to obtain simulation execution results. Based on the simulation execution results, the target port train scheduling scheme is selected from the qualified port train operation collaborative scheduling schemes.

[0158] Specifically, the steps for performing scheduling simulations on port train operation collaborative scheduling schemes according to priority, calling multiple models for parallel simulation and recording data, and selecting at least one qualified port train operation collaborative scheduling scheme after automatic verification by dual baselines can be as follows:

[0159] The above-output scheme clusters are enqueued according to the dual keywords of priority and release time, and scheduled for simulation in sequence. For each scheme, the train operation model, track occupancy model and crane operation model are called for parallel simulation, and the train passing time, track occupancy trajectory and equipment load curve are written into the sandbox status table in real time.

[0160] Using lower limit of throughput With safety limit Two baselines are used to automatically verify the three types of outputs, and the result is determined if and only if all three indicators fall within the range of ≥ and ≤ When the interval is reached, the solution is marked as feasible, and the remaining buffer duration and peak device utilization are written into the solution metadata for subsequent execution.

[0161] The steps for simulating the execution of qualified train operation coordination scheduling schemes at various ports and obtaining the simulation results can be as follows:

[0162] The first feasible solution is executed by encapsulating instructions into a package that sequentially writes three types of information—train entry / exit sequence, track locking period, and crane operation window—into an instruction packet, categorized by resource number, action type, and start / end time. This packet is then pushed to the dispatch console, locomotive control terminal, and yard control terminal via the message bus. A rollback pointer is generated at the database transaction layer, recording the current resource status, command sequence number, and timestamp, ensuring that any subsequent unexpected rollback can be undone with a single pointer. Once the instruction packet reaches each terminal, the field PLC automatically switches to execution mode, completing the solution's deployment.

[0163] The steps for selecting the target port train scheduling scheme from the qualified coordinated scheduling schemes for train operations at various ports based on the simulation results can be as follows:

[0164] After the instruction package is executed on-site, the deviation monitoring module is activated to write the actual train arrival time, loading and unloading progress, and track release progress into the monitoring buffer, and compares them with the baseline of the plan at a rolling interval of 1 second; when any deviation exceeds the threshold... The alarm channel is immediately triggered, and the deviation information is combined with the current resource status to form the latest scene fingerprint. ; Synchronously Push it back to the entry queue of step S3 to drive the subsequent prediction, scheduling, and execution loop to start again.

[0165] The aforementioned port train scheduling method first acquires real-time dynamic scheduling status information for the entire port train operation process; then, it performs graph embedding on the real-time dynamic scheduling status information to obtain a dynamic coupled topology graph for the entire port train operation process; next, it performs threshold edge pruning on the dynamic coupled topology graph to obtain a dynamic coupled local topology graph; and by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, it predicts the occupancy and delay probabilities of each topology node within a preset time window to obtain a joint probability matrix for port train operation resource matching; finally, it uses the joint probability matrix for port train operation resource matching to perform port train operation risk space modeling and safety path mining to obtain the port train... The system first identifies a dynamic feature set of low-risk routes for port train operations. Then, it performs safety-constraint-oriented collaborative scheduling optimization on this feature set to obtain a collaborative scheduling scheme for port train operations. Finally, it selects a target port train scheduling scheme that has passed performance simulation verification from the collaborative scheduling schemes. Based on this, in the dynamic and uncertain port train scheduling environment, the system can link short-term risk prediction results with the scheduling decision-making process in real time, directly outputting a target port train scheduling scheme with minimal conflicts and no need for frequent manual ticket changes. This allows for continuous matching of the on-site rhythm and stable throughput within a short timescale, thus improving the effectiveness of port train scheduling.

[0166] In one feasible approach, refer to Figure 2 , Figure 2This is a schematic diagram of the overall process for scheduling port trains. The overall process may include the following: 1) S1: Receive three types of asynchronous information: terminal loading and unloading plan, train operation instructions, and loading and unloading equipment status. Write each status change into a millisecond-level event queue and update the rolling status tables of trains, tracks, and equipment accordingly; 2) S2: Based on the status tables, construct a physical-temporal dual-layer node topology within a unified field reference system. Perform variable-length flexible segmentation on the tracks and yard operation areas, set loading and unloading coupling nodes and fault gate control nodes, and generate an integrated topology map with dynamic probability weights; 3) S3: Perform threshold edge pruning on the topology map to obtain local subgraphs. Use a spatiotemporal graph convolutional network to predict the occupancy probability and delay probability of each node within a preset time window, and output the train-track joint probability matrix; 4) S4: Combine the joint... The probability matrix is ​​mapped to a three-dimensional risk grid field. High-risk blocks are formed by morphological dilation and equipment shutdown convexity. Low-risk corridors that meet the minimum safety distance are extracted based on distance transformation, and the remaining capacity and risk gradient of the corridor are recorded. 5) S5. The action space is limited by low-risk corridors. The train-track-time slice scheduling scheme is generated by reinforcement learning according to the two-level diffusion strategy of railway layer and loading and unloading layer. During the diffusion process, risk temperature scaling noise and scene memory similarity adjustment are used to achieve fast convergence. Gradient-driven local repair is performed on the atomic actions that fall into high-risk cells. 6) S6. The scheduling scheme is verified by parallel simulation. When it meets the throughput and safety threshold, it is encapsulated into a scheduling instruction package and sent to the dispatching console, locomotive and yard terminal. After the instruction is executed, the field deviation is continuously monitored. When the deviation exceeds the threshold, the backflow is triggered to step S3 for recalculation.

[0167] Reference Figure 3 , Figure 3 This diagram illustrates a comparison between the algorithm used in the port train scheduling method provided in this embodiment and a traditional algorithm. The comparison shows that the algorithm used in this embodiment is significantly more effective.

[0168] Understandably, the system first uses millisecond-level event streams to drive state updates, mapping trains, tracks, and yard equipment to physical-temporal dual-layer nodes under a unified reference frame. Variable-length flexible segmentation is used for tracks, and cross-domain interlocks are modeled through coupling nodes and gating nodes. Subsequently, occupancy, delay, and interlock probabilities are written into edge attributes and overwritten in real time, enabling the topology itself to carry dynamic risk weights. This encapsulates short-term risks and schedulable resources in a unified manner, providing explicit boundaries for subsequent inference without relying on traditional penalty terms. Therefore, it improves the effectiveness of port train scheduling.

[0169] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0170] Based on the same inventive concept, this application also provides a port train scheduling device for implementing the port train scheduling method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more port train scheduling device embodiments provided below can be found in the limitations of the port train scheduling method described above, and will not be repeated here.

[0171] In one exemplary embodiment, such as Figure 4 As shown, a port train dispatching device is provided, comprising: an acquisition module 401, a graph embedding module 402, a prediction module 403, a mining module 404, a dispatching optimization module 405, and a selection module 406, wherein:

[0172] The acquisition module 401 is used to acquire real-time dynamic scheduling status information of the entire port train operation process;

[0173] The graph embedding module 402 is used to embed the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph of the entire port train operation process.

[0174] The prediction module 403 is used to obtain a dynamic coupled local topology graph by performing threshold clipping on the dynamic coupled topology graph, and to predict the occupancy and delay probabilities of each topology node within a preset time window by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, thereby obtaining a joint probability matrix for port train operation resource matching.

[0175] The mining module 404 is used to perform port train operation risk space modeling and safety path mining by matching port train operation resources with a joint probability matrix, and to obtain a dynamic feature set of low-risk paths for port train operations.

[0176] The scheduling optimization module 405 is used to perform safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk routes for port train operations, and obtain a collaborative scheduling scheme for port train operations.

[0177] Module 406 is selected to select a target port train scheduling scheme that has passed the performance simulation verification from the port train operation collaborative scheduling scheme.

[0178] In one embodiment, the acquisition module 401 is further configured to:

[0179] The system acquires real-time status information from asynchronous inputs throughout the entire port train operation process, and synchronizes this real-time status information to a millisecond-level event sequence using a unified event identifier, resulting in a millisecond-level ordered event sequence. This ordered event sequence is then expanded according to the subject dimension and written into a preset event rolling status table to obtain the rolling status information of the port train operation object and the constraint factors affecting the port train operation. Based on the rolling status information and constraint factors, a real-time status vector for the entire port train operation process is generated. Finally, the real-time status information is mapped to the scheduling base information flow to obtain real-time dynamic scheduling status information.

[0180] In one embodiment, the graph embedding module 402 is further configured to:

[0181] Physical nodes are allocated to the job resource objects corresponding to the real-time dynamic scheduling status information, and each physical node is time-replicated according to a preset time window to obtain a two-layer node mapping feature with physical and temporal attributes. Based on the two-layer node mapping feature, the job resource objects are partitioned according to a flexible segmentation strategy to obtain the flexible nodes corresponding to the port trains, and the set of occupied flexible segments of the flexible nodes is matched in a coverage minimization manner. The corresponding blocking weights are matched to the job dynamic constraint information in the set of occupied flexible segments to obtain the structured topology constraint information of the port trains. According to the structured topology constraint information, the reachability combination of all job nodes in the set of occupied flexible segments is updated to obtain the set of job node edges. By performing short-term event statistics on the set of job node edges, the probability vector of job node edges of the port trains is obtained. According to the probability vector of job node edges, a dynamic coupled topology graph is generated.

[0182] In one embodiment, the prediction module 403 is further configured to:

[0183] The dynamic coupling topology graph is input into a preset edge threshold layer to obtain the scalar values ​​of the workload and conflict weight of each directed edge in the dynamic coupling topology graph. All workload and conflict weight scalar values ​​are identified by a preset dynamic threshold function, and the available directed edges of all directed edges in the dynamic coupling topology graph are identified. By merging all available directed edges in the same direction, a dynamic coupling local topology graph is obtained.

[0184] In one embodiment, the mining module 404 is further configured to:

[0185] The joint probability matrix is ​​decomposed line by line according to the predetermined route of the port train to obtain the operational constraint probability of the port train under each route. Based on the operational constraint probability and the corresponding on-site load weight, the comprehensive operational risk value of the port train under each route is generated. The path containing the comprehensive operational risk value is subjected to 3D raster mapping and interpolation completion processing to obtain the 3D continuous risk field corresponding to the port train. The 3D continuous risk field is subjected to risk region hierarchical fusion processing to obtain the high-risk block set of the port train. Risk block constraint extension processing is performed in the high-risk block set to obtain the safety buffer zone under the hard constraint of the operational resource object. The passable grid cells of the port train are selected from the safety buffer zone, and the risk perception scheduling boundary of the port train is extracted based on the shortest path search of the passable grid cells via the time layer. Based on the risk perception boundary, the remaining available time, remaining capacity and risk gradient information of the low-risk path are standardized to obtain the dynamic feature set of the low-risk path of the port train operation.

[0186] In one embodiment, the scheduling optimization module 405 is further configured to:

[0187] A two-layer action set between the railway layer and the loading / unloading layer is generated based on the dynamic feature set of the risk path. A traceable scheduling skeleton is constructed based on the inter-layer mapping relationship established by the two-layer action set. Based on the traceable scheduling skeleton, priority-oriented initial diffusion noise is assigned to the railway layer actions, resulting in the convergent train sequence for the railway layer and prior constraints for the loading / unloading layer actions. Under the convergent train sequence for the railway layer and the prior constraints for the loading / unloading layer actions, the three-dimensional comprehensive risk field is mapped to a temperature tensor. The amplitude of the action diffusion noise is dynamically adjusted through a temperature scaling model, resulting in a reinforcement learning safety exploration rule embedded in the safety boundary. Then, based on the three types of features of the diffusion mid-stage scene extracted by the reinforcement learning safety exploration rules, the scene fingerprint is spliced ​​to obtain the scene fingerprint. After matching the scene fingerprint with the historical memory database, the diffusion noise center of this round is updated according to the segment replacement rule. After the diffusion converges in the railway layer and the loading and unloading layer, the risk field back lookup is performed on the output port train operation collaborative scheduling scheme to obtain the risk field back lookup results. The risk field back lookup results include at least one candidate port train operation collaborative scheduling scheme. Among the candidate port train operation collaborative scheduling schemes, the port train operation collaborative scheduling scheme is selected according to business preferences.

[0188] In one embodiment, the selection module 406 is further configured to:

[0189] The port train operation collaborative scheduling schemes are simulated one by one according to priority. Multiple models are called for parallel simulation and data is recorded. After automatic verification by dual baselines, at least one qualified port train operation collaborative scheduling scheme is selected. The qualified port train operation collaborative scheduling schemes are simulated and executed to obtain simulation execution results. Based on the simulation execution results, the target port train scheduling scheme is selected from the qualified port train operation collaborative scheduling schemes.

[0190] Each module in the aforementioned port train dispatching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0191] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a port train scheduling method. Those skilled in the art will understand that... Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0192] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0193] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0194] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0195] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0196] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0197] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A port train scheduling method, characterized in that, The method includes: Obtain real-time dynamic scheduling status information for the entire port train operation process; The real-time dynamic scheduling status information is embedded into a graph to obtain a dynamic coupled topology graph of the entire port train operation process; By performing threshold clipping on the dynamic coupled topology graph, a dynamic coupled local topology graph is obtained. Then, by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, the occupancy and delay probabilities of each topology node within a preset time window are predicted to obtain a joint probability matrix for port train operation resource matching. By matching the port train operation resources with the joint probability matrix, the port train operation risk space modeling and safe path mining are performed to obtain the dynamic feature set of low-risk paths for port train operations. A collaborative scheduling optimization based on safety constraints is performed on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations. Select the target port train scheduling scheme that has passed the performance simulation verification from the aforementioned port train operation collaborative scheduling scheme.

2. The method according to claim 1, characterized in that, The acquisition of real-time dynamic scheduling status information for the entire port train operation process includes: The real-time status information of the asynchronous input of the entire port train operation process is obtained, and the real-time status information is synchronized to the millisecond-level event sequence according to the unified event identifier to obtain the millisecond-level ordered event sequence; The millisecond-level ordered event sequence is expanded according to the subject dimension and written into a preset event rolling state table to obtain the rolling state information of the port train operation object and the constraint factor information affecting the port train operation. Based on the rolling state information and the constraint factor information, a real-time state vector of the entire port train operation process is generated. The real-time status information is mapped to the scheduling base information stream to obtain the real-time dynamic scheduling status information.

3. The method according to claim 1, characterized in that, The step of embedding the real-time dynamic scheduling status information into a graph to obtain a dynamically coupled topology graph of the entire port train operation process includes: Physical nodes are allocated to the job resource objects corresponding to the real-time dynamic scheduling status information, and time replication is performed on each physical node according to a preset time window to obtain a two-layer node mapping feature with physical attributes and time attributes. Based on the dual-layer node mapping features, the operation resource object is divided into flexible nodes corresponding to the port train according to the flexible segmentation strategy, and the set of occupied flexible segments of the flexible nodes is matched in the manner of coverage minimization. Match the corresponding blocking weights to the operational dynamic constraint information in the set of occupied flexible sections to obtain the structured topology constraint information of the port train; Based on the structured topology constraint information, update the reachability combination of all job nodes in the set of occupied flexible segments to obtain the set of job node edges; By performing short-term event statistics on the set of operation node edges, the probability vector of the operation node edges of the port train is obtained; The dynamic coupled topology graph is generated based on the edge probability vector of the job node.

4. The method according to claim 1, characterized in that, The process involves performing threshold edge pruning on the dynamically coupled topology graph to obtain a dynamically coupled local topology graph, and then inputting the dynamically coupled local topology graph into a spatiotemporal graph convolutional network model to predict the occupancy and delay probabilities of each topology node within a preset time window, thereby obtaining a joint probability matrix for port train operation resource matching. This includes: The dynamic coupling topology graph is input to a preset edge threshold layer to obtain the scalar values ​​of the workload and conflict weight of each directed edge in the dynamic coupling topology graph. By identifying all job load scalar values ​​and conflict weight scalar values ​​through a preset dynamic threshold function, the available directed edges of all directed edges in the dynamic coupling topology graph are identified. The dynamically coupled local topology graph is obtained by merging all available directed edges in the same direction.

5. The method according to claim 1, characterized in that, The process involves matching the port train operation resources with the joint probability matrix to perform port train operation risk space modeling and safe path mining, resulting in a dynamic feature set of low-risk paths for port train operations, including: The joint probability matrix is ​​decomposed line by line according to the predetermined route of the port train to obtain the operational constraint probability of the port train under each route, and the comprehensive operational risk value of the port train under each route is generated based on the operational constraint probability and the corresponding on-site load weight. The path containing the comprehensive risk value of the operation is subjected to three-dimensional rasterization mapping and interpolation completion processing to obtain the three-dimensional continuous risk field corresponding to the port train; The three-dimensional continuous risk field is subjected to risk region hierarchical fusion processing to obtain the high-risk block set of the port train; The risk block constraint extension process is performed on the high-risk block set to obtain a safety buffer zone under the hard constraint of the job resource object. The passable grid cells of the port train are selected from the safety buffer zone, and then the risk perception scheduling boundary of the port train is extracted based on the shortest path search with time layer of the passable grid cells. Based on the risk perception boundary, the remaining available time, remaining capacity, and risk gradient information of the low-risk path are standardized to obtain the dynamic feature set of the low-risk path for port train operations.

6. The method according to claim 1, characterized in that, The process of performing safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations includes: Based on the dynamic feature set of the risk path, a two-layer action set between the railway layer and the loading and unloading layer is generated, and a traceable scheduling skeleton is constructed according to the inter-layer mapping relationship established by the two-layer action set. Based on the traceable scheduling skeleton, the initial diffused noise is allocated to the railway layer action priority guidance to obtain the railway layer convergent train sequence and the loading and unloading layer action prior constraints. Under the prior constraints of the converged train sequence in the railway layer and the actions in the loading and unloading layer, the three-dimensional comprehensive risk field is mapped into a temperature tensor. The amplitude of the action diffusion noise is dynamically adjusted through the temperature scaling model to obtain the reinforcement learning safety exploration rules embedded in the safety boundary. Based on the three types of features of the diffusion mid-stage scene extracted by the reinforcement learning security exploration rules, a scene fingerprint is obtained by splicing them together. After matching the scene fingerprint with the historical memory database, the diffusion noise center of this round is updated according to the segment replacement rules. After the diffusion convergence of the railway layer and the loading and unloading layer, a risk field back lookup is performed on the output port train operation collaborative scheduling scheme to obtain the risk field back lookup results, wherein the risk field back lookup results include at least one candidate port train operation collaborative scheduling scheme. The port train operation collaborative scheduling scheme is selected from the candidate port train operation collaborative scheduling schemes according to business preferences.

7. The method according to claim 1, characterized in that, The selection of a target port train scheduling scheme that has passed performance simulation verification from the port train operation collaborative scheduling scheme includes: The port train operation collaborative scheduling scheme is simulated in one step according to priority, multiple models are called for parallel simulation and data is recorded. After automatic verification by dual baselines, at least one qualified port train operation collaborative scheduling scheme is selected. The qualified scheduling schemes for port train operations were simulated and the simulation results were obtained. Based on the simulation results, the target port train scheduling scheme is selected from the qualified port train operation coordination scheduling schemes.

8. A port train dispatching device, characterized in that, The device includes: The acquisition module is used to acquire real-time dynamic scheduling status information of the entire port train operation process; The graph embedding module is used to embed the real-time dynamic scheduling status information into a graph to obtain a dynamic coupled topology graph of the entire port train operation process. The prediction module is used to obtain a dynamic coupled local topology graph by performing threshold clipping on the dynamic coupled topology graph, and to predict the occupancy and delay probabilities of each topology node within a preset time window by inputting the dynamic coupled local topology graph into a spatiotemporal graph convolutional network model, thereby obtaining a joint probability matrix for port train operation resource matching. The mining module is used to perform risk space modeling and safe passage mining of port train operations by matching the port train operation resources with the joint probability matrix, and to obtain a dynamic feature set of low-risk passages for port train operations. The scheduling optimization module is used to perform safety-constraint-oriented collaborative scheduling optimization on the dynamic feature set of low-risk routes for port train operations to obtain a collaborative scheduling scheme for port train operations. The selection module is used to select a target port train scheduling scheme that has passed the performance simulation verification from the port train operation collaborative scheduling scheme.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images