Container scheduling management method and system based on NP-LSTM double-branch fusion

Abstract

The application provides a container scheduling management method and system based on NP-LSTM double-branch fusion, relates to the technical field of data processing, and obtains container task data, yard resource data and operation equipment state data of a target voyage and forms an operation combination; time operation data is processed through an LSTM branch, and a berth space constraint term is processed through an NP branch, so that a time distribution value and a berth constraint value are obtained; a future container lifting conflict chain is generated in combination with a ship stowage sequence, upper and lower layer shielding relationships and the berth space constraint term; an initial scheduling value is obtained according to double-branch weights, a conflict correction amount is determined by using a conflict type, a conflict node quantity and a directed correlation order, and a front conflict resolution value is obtained; the target operation combination and the target operation sequence are determined according to the conflict correction amount and the front conflict resolution value, and container scheduling instructions are output.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a container scheduling management method and system based on NP-LSTM dual-branch fusion. Background Technology

[0002] Existing container dispatching and management technologies typically target container entry, exit, storage, loading, unloading, and transshipment tasks in ports, railway stations, or multimodal transport yards. They collect operational data such as ship berthing schedules, train arrival and departure schedules, container numbers, container types, container weights, destination ports, yard bay positions, operating machinery status, and truck or AGV locations. The system generally uses rule constraints, single prediction models, or traditional recurrent neural networks to predict container arrival volumes, operation durations, and equipment occupancy status. Then, based on customs clearance times, yard distances, machinery availability, and task priorities, it generates dispatching instructions to allocate containers among quay cranes, yard cranes, yard blocks, and transport vehicles.

[0003] However, in real-world scenarios where export containers arrive at ports in concentrated quantities and require rapid loading, existing technologies may exhibit technical flaws where scheduling results do not align with actual yard operational constraints. For example, consider the operations of a coastal container terminal before the cut-off time for a particular foreign trade route. Ordinary containers, refrigerated containers, dangerous goods containers, and oversized containers from the same voyage may enter the yard through the gates in a short period. Existing scheduling systems, relying primarily on time-series predictions for a single path, typically focus on vehicle arrival times, equipment idle times, and estimated operation durations, treating container flow changes as a continuous time series. However, in actual yards, different container types are simultaneously constrained by spatial location, stacking rules, power supply resources, safety isolation distances, and ship loading sequences. For instance, refrigerated containers must be allocated to areas with sufficient power supply for cold-plugging, dangerous goods containers need to enter dedicated areas meeting isolation requirements, and oversized containers cannot be simply stacked in ordinary container positions. Furthermore, containers prioritized for loading later in the same voyage should not be placed on lower levels or in areas far from the loading line. If the system allocates containers to seemingly idle general bays solely based on predicted arrival times and current equipment load, a seemingly executable stacking plan may be formed during the arrival phase. However, when retrieving containers according to the ship's stowage plan before loading, it may be discovered that some bays do not meet the requirements for refrigeration, power supply, hazardous materials isolation, or priority pick-up routes, necessitating temporary container handling, relocation, or inter-regional transfer. This deficiency is not simply manifested as low scheduling efficiency or high operating costs, but rather as the current technology's failure to simultaneously distinguish between "time-series fluctuation characteristics" and "nonlinear spatial constraint characteristics" leads to a deviation between the predicted results and the actual executable yard constraints. This causes conflicts between front-end arrival allocation and back-end loading operations, affecting the accuracy and continuous execution of container scheduling instructions. Summary of the Invention

[0004] The purpose of this invention is to provide a container scheduling management method and system based on NP-LSTM dual-branch fusion, aiming to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a container scheduling management method based on NP-LSTM dual-branch fusion, the method comprising:

[0007] Acquire container mission data, yard resource data, and operational equipment status data for the target voyage to form an initial dataset;

[0008] Based on the initial dataset, the containers to be allocated, candidate bay locations, and available operating equipment are determined, and operating combinations are formed.

[0009] Based on the time-based task data corresponding to the task combination, a time input quantity is generated to characterize the fluctuation of the task time series and input into the LSTM branch to obtain the time allocation value.

[0010] Based on the space constraint term of the candidate spaces in the job combination, a spatial position input quantity to characterize the nonlinear spatial constraint is generated and input into the NP branch to obtain the space constraint value.

[0011] Based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination, a future container lifting conflict chain is generated, consisting of conflict nodes and directed associations between conflict nodes.

[0012] The time allocation values ​​and bit constraint values ​​of the same job combination are mapped to the normalized evaluation interval, and the LSTM branch weights and NP branch weights are determined respectively. The weighted values ​​are then used to obtain the initial scheduling values.

[0013] The conflict correction amount is determined based on the conflict type, number of conflict nodes and directed association level in the future container picking conflict chain, and the initial scheduling value is corrected to obtain the previous conflict resolution value.

[0014] Based on the conflict correction amount, candidate target operation combinations corresponding to each container to be allocated are selected, and the target operation combination and target operation sequence are determined from the candidate target operation combinations based on the prior conflict resolution value, and the container scheduling instruction is output.

[0015] Secondly, a container scheduling and management system based on NP-LSTM dual-branch fusion, the system comprising:

[0016] The data acquisition module is used to acquire container task data, yard resource data, and operating equipment status data for the target voyage to form an initial dataset;

[0017] The job combination generation module is used to determine the containers to be allocated, candidate bays, and available operating equipment based on the initial data set, and to form job combinations.

[0018] The LSTM branch processing module is used to generate time input quantities that characterize the fluctuation of the job time series based on the time job data corresponding to the job combination, and input them into the LSTM branch to obtain the time allocation value.

[0019] The NP branch processing module is used to generate spatial location inputs that characterize nonlinear spatial constraints based on the space constraint terms of candidate spaces in the job combination, and input them into the NP branch to obtain the space constraint values.

[0020] The conflict chain generation module is used to generate a future container loading conflict chain based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination.

[0021] The dual-branch fusion module is used to map the time allocation value and bit constraint value of the same job combination to the normalized evaluation interval, and determine the LSTM branch weight and NP branch weight respectively, and perform weighted summation to obtain the initial scheduling value.

[0022] The conflict correction module is used to determine the conflict correction amount based on the conflict type, number of conflict nodes and directed association level in the future container lifting conflict chain, and to correct the initial scheduling value to obtain the previous conflict resolution value.

[0023] The scheduling instruction output module is used to filter candidate target operation combinations corresponding to each container to be allocated based on the conflict correction amount, and determine the target operation combination and target operation sequence from the candidate target operation combinations based on the prior conflict resolution value, and output the container scheduling instruction.

[0024] The above-described solution of the present invention has at least the following beneficial effects:

[0025] This invention combines containers to be allocated, candidate bays, and available operating equipment into a single operation group. This ensures that arrival time, equipment status, bay conditions, and subsequent scheduling outputs are all processed around the same operation object, avoiding the problem of inconsistency between the time evaluation object and the yard constraint object when generating scheduling results solely based on container arrival time or equipment idle status.

[0026] Furthermore, this invention uses LSTM branching to characterize the time input of the operation time series fluctuation and NP branching to characterize the spatial position input of the nonlinear spatial constraints. This allows time factors such as the expected arrival time of the gate, the remaining time of the gate closure, and the expected loading and unloading time to be evaluated separately from spatial factors such as the layer, adjacent occupancy, cold plug power supply, safety isolation, over-limit stacking, and the distance of the operation channel. These factors are then fused together to reduce the deviation caused by the single path time series prediction results on the bay allocation.

[0027] Furthermore, this invention generates a future container loading conflict chain based on the ship loading sequence, the upper and lower layer obstruction relationship, and the bay space constraint. It transforms subsequent operational problems such as insufficient cold plug power supply, insufficient safety isolation, insufficient over-limit stacking, restricted operation channels, and obstruction during loading and unloading into conflict nodes and directed associations, so that the stacking position selection in the entry stage can reflect the obstruction relationship in the loading and unloading stage.

[0028] Furthermore, the present invention determines the conflict correction amount based on the conflict type, number of conflict nodes, and directed association level in the future container requisition conflict chain, and uses the conflict correction amount to correct the initial scheduling value, so that the operation combination containing multiple types of conflicts, multiple conflict nodes, or multiple levels of directed association are corrected accordingly when selecting the target, thereby reducing the need for container flipping, moving, or cross-regional transfer of surface vacant bays before subsequent loading due to power supply, isolation, over-limit stacking, or container obstruction.

[0029] Finally, the present invention filters candidate target operation combinations based on conflict correction amount, and determines target operation combinations and target operation sequences based on prior conflict resolution value, so that the output container scheduling instructions are simultaneously constrained by time operation status, bay space constraints and future container pick-up conflicts, thereby improving the consistency between front-end arrival allocation and back-end loading operations. Attached Figure Description

[0030] Figure 1 This is a flowchart of a container scheduling management method based on NP-LSTM dual-branch fusion provided by an embodiment of the present invention.

[0031] Figure 2 This is a flowchart illustrating the generation of future box-picking conflict chains provided by an embodiment of the present invention.

[0032] Figure 3 This is a flowchart illustrating the determination of LSTM branch weights and NP branch weights provided in an embodiment of the present invention. Detailed Implementation

[0033] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0034] In this invention, NP stands for Neighbor Position, which is used to characterize the relationship between candidate bays and surrounding spatial locations during container scheduling. The surrounding spatial locations include the layer of the candidate bay, the occupancy status of adjacent bays, the cold plug power supply distance, the safety isolation distance, the over-limit stacking limit, and the operating channel distance.

[0035] In this invention, LSTM stands for Long Short-Term Memory, which is used to characterize the recursive memory processing related to changes in operation time during container scheduling. The changes in operation time include the estimated arrival time at the gate, the remaining time before the gate is closed, the estimated loading and unloading time, the idle time of available operating equipment, and the time series fluctuations formed by the changes in the distance from available operating equipment to candidate bays as the operation combination changes.

[0036] In this invention, NP-LSTM represents a dual-branch fusion structure in which time series fluctuation characteristics and nonlinear spatial constraint characteristics jointly participate in scheduling decisions. The NP branch is responsible for spatial constraint processing, and the LSTM branch is responsible for time series processing. The outputs of the two branches jointly participate in the calculation of scheduling values.

[0037] like Figure 1 As shown, embodiments of the present invention propose a container scheduling management method based on NP-LSTM dual-branch fusion, the method comprising:

[0038] Acquire container mission data, yard resource data, and operational equipment status data for the target voyage to form an initial dataset;

[0039] Based on the initial dataset, the containers to be allocated, candidate bay locations, and available operating equipment are determined, and operating combinations are formed.

[0040] Based on the time-based operation data corresponding to the job combination, a time input quantity is generated to characterize the fluctuation of the job time series and input into the LSTM branch to obtain the time allocation value. The time-based operation data includes the expected arrival time of the gate, the remaining time of the gate closure, the expected loading and unloading time, the idle time of the available operation equipment, and the distance from the available operation equipment to the candidate bay. The time allocation value is used to characterize the executability of the job combination in the time dimension.

[0041] Based on the bayonet space constraint terms of candidate bayonet positions in the job combination, a spatial position input quantity for characterizing nonlinear spatial constraints is generated and input into the NP branch to obtain the bayonet constraint value; the bayonet space constraint terms include layer position constraint terms, adjacent position occupancy constraint terms, cold plug power supply constraint terms, safety isolation constraint terms, over-limit stacking constraint terms, and job channel distance constraint terms; the bayonet constraint value is used to characterize the degree to which the job combination meets the bayonet space constraint terms;

[0042] Based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination, a future container lifting conflict chain is generated, consisting of conflict nodes and directed associations between conflict nodes.

[0043] The time allocation values ​​and B-value constraints of the same job combination are mapped to the normalized evaluation interval, and the LSTM branch weights and NP branch weights are determined respectively. The weighted average is then used to obtain the initial scheduling value. The initial scheduling value is used to comprehensively evaluate the scheduling priority after considering both the time and spatial dimensions.

[0044] The conflict correction amount is determined based on the conflict type, number of conflict nodes, and directed association level in the future container pick-up conflict chain, and the initial scheduling value is corrected to obtain the pre-conflict resolution value; the pre-conflict resolution value is used to characterize the final scheduling priority after considering future container pick-up conflicts.

[0045] Based on the conflict correction amount, candidate target operation combinations corresponding to each container to be allocated are selected, and the target operation combination and target operation sequence are determined from the candidate target operation combinations based on the prior conflict resolution value, and the container scheduling instruction is output.

[0046] In this embodiment of the invention, the method combines the container to be allocated, the candidate bay space, and the available operating equipment into the same operating combination, so that the time operation data, bay space constraints, and subsequent scheduling output are all processed around the same object, avoiding the problem of inconsistency between the time evaluation object and the space evaluation object when judging only the container or bay space as a separate object.

[0047] This method uses time input to characterize the fluctuation of the operation time series and spatial location input to characterize the nonlinear spatial constraints of candidate bay positions. It separates time factors such as the expected arrival time of the gate, the remaining time of the gate closure, the expected loading and unloading time, and the idle time of the equipment from spatial factors such as the layer, adjacent occupancy, cold plug power supply, safety isolation, over-limit stacking, and the distance of the operation channel, and then performs a fusion evaluation, thereby reducing the bay position allocation deviation caused by the mixed processing of time factors and spatial constraints.

[0048] Before generating scheduling results, this method combines the ship loading sequence, the upper and lower layer obstruction relationship, and bay space constraints to form a future container retrieval conflict chain. It transforms the obstruction, power supply, safety isolation, over-limit stacking, and channel occupation conflicts that occur during the loading and retrieval stage into conflict nodes and directed associations in advance, so that the bay allocation during the entry stage can reflect the obstruction relationship in the subsequent loading and retrieval process.

[0049] This method determines the conflict correction amount based on the conflict type, number of conflict nodes, and directed association level in the future container pickup conflict chain, and uses the conflict correction amount to correct the initial scheduling value. This reduces the probability of using operation combinations with multi-level obstruction, unmet hard space constraints, or channel occupancy conflicts in the target operation combination screening, thereby reducing the number of container scheduling instructions that are time-executable but blocked during the loading and pickup stage.

[0050] This method selects candidate target operation combinations based on conflict correction values ​​and determines the target operation combinations and target operation sequences based on the prior conflict resolution values. This ensures that the target bay, target operation equipment, and target operation sequence are simultaneously constrained by the time operation status, bay space constraints, and future container retrieval conflicts, thereby reducing temporary container relocation, overturning, and cross-area adjustments caused before loading due to bay obstruction, insufficient cold plug power supply, insufficient safety isolation, insufficient over-limit stacking, or limited operation channels.

[0051] In a preferred embodiment of the present invention, container task data, yard resource data, and operating equipment status data for the target voyage are acquired to form an initial data set, including:

[0052] Based on the target voyage identifier, the container task data corresponding to the target voyage is read from the terminal business database. The container task data includes container number, container type, refrigerated label, dangerous goods label, oversized label, estimated gate arrival time, remaining time before customs clearance, estimated loading and unloading time, and vessel loading sequence.

[0053] Read the yard resource data corresponding to the target voyage from the yard resource database. The yard resource data includes container area, candidate bay number, layer, bay occupancy status, adjacent bay occupancy status, cold plug power supply point location, safety isolation area location, over-limit stacking restrictions, and operation channel location.

[0054] Read the status data of the work equipment from the work equipment management system. The status data of the work equipment includes the work equipment number, work equipment type, current location of the work equipment, idle time of the work equipment, and availability status of the work equipment.

[0055] The container mission data, yard resource data, and operating equipment status data are linked according to the target voyage identifier to form an initial data set.

[0056] In a preferred embodiment of the present invention, based on an initial data set, containers to be allocated, candidate bay locations, and available operating equipment are determined, and operating combinations are formed, including:

[0057] Extract containers from the initial dataset that have not yet been allocated to the yard for the target voyage and identify them as containers to be allocated.

[0058] Extract the bays that are available for allocation from the initial dataset, and determine the candidate bays corresponding to the containers to be allocated based on the container type, refrigerated label, dangerous goods label, and oversize label.

[0059] Extract the available operating equipment from the initial dataset, and determine the available operating equipment corresponding to the containers to be allocated and the candidate bay locations based on the type of operating equipment, the current location of the operating equipment, and the idle time of the operating equipment;

[0060] A task combination is formed by combining the same container to be allocated, a candidate bay, and an available operating equipment, and a task combination identifier is established for the task combination.

[0061] In a preferred embodiment of the present invention, let the first... The task combination is as follows The normalized evaluation interval is [ , ],in, This includes a container to be allocated, a candidate bay, and an available operating device, wherein, This is the preset lower limit value of the normalized evaluation interval. This is the preset upper limit of the normalized evaluation interval, and < .

[0062] The time allocation values ​​are formed as follows:

[0063] set up The corresponding gate's estimated arrival time is The available idle time of the working equipment is The distance from the available working equipment to the candidate bay location is The preset running speed is The estimated loading and unloading time is The remaining time before the deadline is The scheduling reference time is ;

[0064] The expected start time is: ;

[0065] The travel time is: ;

[0066] The estimated completion time is: ;

[0067] The cutoff time is: ;

[0068] The time margin is: ;

[0069] Will , , , , , , , and After standardization, the time input quantities are composed in chronological order. Time input Represented as: ;

[0070] Input time Each time term is input into the LSTM branch in chronological order, and the hidden state of the current time term is updated according to the hidden state of the previous time term to represent the time series fluctuations corresponding to the job combination. The update process is as follows: ; ; ; ; ; ;

[0071] in, For the input gate state, In the forget gate state, Output gate state Candidate memory states, For memory state, This is an implicit output state. , , , , , , , , , , and For LSTM branch parameters;

[0072] When the LSTM branch reads to the last time item Then, based on the last implicit output state Get time allocation value : ;

[0073] in, and These are the parameters for the time output layer.

[0074] In a preferred embodiment of the present invention, the NP branch reads the spatial position input item by item in the order of layer constraint item, adjacent occupancy constraint item, cold plug power supply constraint item, safety isolation constraint item, over-limit stacking constraint item and working channel distance constraint item; for each bay space constraint item read, the spatial state of the current bay space constraint item is updated in combination with the spatial state corresponding to the previous bay space constraint item, until the bay constraint value is output after all bay space constraint items have been updated.

[0075] The beta constraint values ​​are formed as follows:

[0076] set up The corresponding number of beta space constraint terms is The bay space constraints include layer constraints, adjacent occupancy constraints, cold plug power supply constraints, safety isolation constraints, over-limit stacking constraints, and work passage distance constraints. Each bay space constraint item sets the constraint judgment value. ;

[0077] The corresponding layer position, occupancy status, cold plug power supply distance, safety isolation distance, over-limit stacking limit, working channel distance, and constraint judgment value for each bay space constraint item are defined. After standardization, it forms the spatial location input quantity. Spatial location input Represented as: ;

[0078] in, Indicates the first Spatial input terms corresponding to each bay space constraint term;

[0079] Input spatial location Input the NP branch, which is read sequentially according to the order of the Bébit space constraint terms. And update according to the following process: ; ;

[0080] in, For the first Spatial correlation values ​​corresponding to each bay space constraint term. For the first The spatial state value corresponding to each bay space constraint term. This represents the spatial state value corresponding to the previous bayonet spatial constraint term. , and For NP branch parameters, For the first The reduction coefficients corresponding to each bay space constraint term, and , For nonlinear spatial correlation mapping functions, in one embodiment, the Sigmoid function can be used; where, This was obtained based on the statistical analysis of the impact of container pickup corresponding to the directed correlation level in historical scheduling data;

[0081] When the When the space constraint terms are satisfied... ,but When the first When the space constraint term is not satisfied ,but This reduces the spatial state value by eliminating unsatisfied B-space constraints.

[0082] When the NP branch reads to the last BIF space constraint term Then, based on the last spatial state value Obtain the beta constraint value : ;

[0083] in, and These are the output layer parameters for the bayonet.

[0084] In a preferred embodiment of the present invention, the LSTM branch parameters and LSTM branch output layer parameters are obtained through training with historical job samples. The historical job samples include time job data corresponding to historical job combinations and time allocation labels corresponding to historical job combinations. The time job data of the historical job combinations is used to generate time input samples, which are then input into the LSTM branch. The LSTM branch output result is compared with the time allocation label value. Based on the comparison result, the LSTM branch parameters and LSTM branch output layer parameters are adjusted until the difference between the output result and the time allocation label value meets a preset training stopping condition, thus obtaining the trained LSTM branch parameters and LSTM branch output layer parameters.

[0085] In a preferred embodiment of the present invention, the NP branch parameters and NP branch output layer parameters are obtained through training with historical job samples. The historical job samples include the bayonet space constraint terms corresponding to the historical job combinations and the bayonet constraint annotation values ​​corresponding to the historical job combinations. The bayonet space constraint terms of the historical job combinations are used to generate spatial location input samples and input into the NP branch. The output results of the NP branch are compared with the bayonet constraint annotation values. The NP branch parameters and NP branch output layer parameters are adjusted according to the comparison results until the difference between the output results and the bayonet constraint annotation values ​​meets the preset training stopping condition, thus obtaining the trained NP branch parameters and NP branch output layer parameters.

[0086] In a preferred embodiment of the present invention, the historical operation sample includes container task data, yard resource data, operation equipment status data, and corresponding actual scheduling results. The time allocation label value is generated based on the actual completion time of the operation. The bay space constraint label value is generated based on whether the corresponding operation combination satisfies all bay space constraints. Mean squared error is used as the loss function for training, and training stops when the validation set loss does not decrease for a preset number of consecutive rounds.

[0087] In a preferred embodiment of the present invention, the normalized evaluation interval mapping formula is as follows:

[0088] For any original value in the same batch of operations Let the batch of operations consist of... The maximum value is The minimum value is The normalized value is: ;

[0089] when When the normalized value is: ;

[0090] Normalized time allocation value Normalized Bayesian constraint value .

[0091] like Figure 2 As shown, in a preferred embodiment of the present invention, the process of generating a future box-picking conflict chain includes:

[0092] The job combination is determined as the starting node, and the stacking status of the containers to be allocated after being placed into the candidate bays is simulated.

[0093] Based on the simulated stacking status, ship loading sequence, and upper and lower layer obstruction relationship, stacked containers that are affected by the placement of containers to be allocated and are obstructed during loading and retrieval are identified as upper and lower layer obstruction conflict nodes; based on the satisfaction of bay space constraints, objects that do not meet the cold insertion power supply constraint, safety isolation constraint, over-limit stacking constraint, or operation channel distance constraint are identified as cold insertion power supply conflict nodes, safety isolation conflict nodes, over-limit stacking conflict nodes, or operation channel conflict nodes, respectively.

[0094] Establish a first-level directed association by pointing from the starting node to each conflicting node;

[0095] The conflict node pointed to by the first-level directed association is taken as the current conflict node. When the current conflict node continues to affect other conflict nodes, a next-level directed association is established from the current conflict node to the affected conflict node, and the affected conflict node is updated to the current conflict node for iteration until there are no more conflict nodes that are still affected. When the container corresponding to the current conflict node is located above the containers corresponding to other conflict nodes, it is determined to be a continued influence.

[0096] When establishing a next-level directed association, if the directed association to be established causes a closed loop between conflicting nodes, then the directed association to be established will not be established.

[0097] Starting from the initial node, conflicting nodes are arranged along the direction of directed association to form at least one acyclic future box-lifting conflict chain. The number of directed associations contained in the longest directed path from the initial node to each conflicting node is determined as the directed association level of the corresponding conflicting node.

[0098] In this embodiment of the invention, the work combination is used as the starting node. After simulating the stacking state of containers to be allocated after being placed in candidate bays, conflict nodes such as upper and lower layer obstruction conflict nodes, cold insertion power supply conflict nodes, safety isolation conflict nodes, over-limit stacking conflict nodes, and work channel conflict nodes are determined respectively. Through the first-level directed association from the starting node to each conflict node, and the next-level directed association formed when a conflict node continues to affect other conflict nodes, the continuous impact of an entry stacking action on the subsequent loading and retrieval process can be expressed. By stopping the establishment of corresponding directed associations when a closed loop is formed, and determining the level of directed association with the longest directed path, the future container retrieval conflict chain can be used for subsequent conflict correction calculations, avoiding the inclusion of mutually cyclical conflict relationships in the scheduling evaluation.

[0099] In a preferred embodiment of the present invention, when the loading sequence of the already stacked container is earlier than that of the container above it, and the container to be allocated is located on the extraction path of the already stacked container after being placed in the candidate bay, or when the number of obstruction layers above the already stacked container is increased, the already stacked container is identified as an upper and lower layer obstruction conflict node.

[0100] In a preferred embodiment of the present invention, objects that do not meet the cold plug-in power supply constraint, safety isolation constraint, over-limit stacking constraint, or operation channel distance constraint include containers to be allocated, candidate bays, container areas to which candidate bays belong, or stacked containers that are spatially associated with candidate bays.

[0101] In a preferred embodiment of the present invention, when the container corresponding to the current conflict node is located above the container corresponding to other conflict nodes, the candidate bay location corresponding to the current conflict node violates the safe isolation distance with the candidate bay locations corresponding to other conflict nodes, or the operating time period corresponding to the current conflict node overlaps with the operating time period corresponding to other conflict nodes and they jointly occupy the same operating channel, it is determined that the current conflict node continues to affect other conflict nodes.

[0102] In a preferred embodiment of the present invention, before establishing the next level of directed association, it is determined whether the affected conflict node already exists in the directed path from the starting node to the current conflict node; if the affected conflict node already exists in the directed path, it is determined that the directed association to be established will form a closed loop between the conflict nodes.

[0103] like Figure 3 As shown, in a preferred embodiment of the present invention, the process of determining the LSTM branch weights and NP branch weights includes:

[0104] The later of the estimated arrival time of the gate corresponding to the work combination and the idle time of the available work equipment is determined as the estimated start time;

[0105] The travel time is determined based on the distance from the available operating equipment to the candidate bay location and the preset operating speed, and the expected completion time is determined based on the expected start time, travel time, and expected loading and unloading time.

[0106] The cut-off time is determined based on the remaining time before the cut-off, and the time margin is determined based on the difference between the cut-off time and the expected completion time.

[0107] The time urgency coefficient is determined based on the preset time interval to which the time margin belongs. The smaller the time margin, the larger the time urgency coefficient.

[0108] Based on the layer constraint, adjacent occupancy constraint, cold plug power supply constraint, safety isolation constraint, over-limit stacking constraint and operation channel distance constraint, determine whether the candidate bays meet the corresponding bay space constraint, and count the number of bay space constraints that do not meet the constraint.

[0109] The spatial conflict coefficient is determined based on the preset number range to which the number of unsatisfied bay space constraints belongs. The more unsatisfied bay space constraints there are, the greater the spatial conflict coefficient will be.

[0110] The sum of the time urgency coefficient and the spatial conflict coefficient is determined as the normalization base. When the normalization base is not zero, the ratios of the time urgency coefficient and the spatial conflict coefficient to the normalization base are calculated to obtain the LSTM branch weights and NP branch weights. When the normalization base is zero, the LSTM branch weights and NP branch weights are determined as preset base weights, where the sum of the LSTM branch weights and NP branch weights is one.

[0111] In this embodiment of the invention, the estimated completion time is first determined based on the gate's expected arrival time, equipment idle time, equipment travel time, and estimated loading / unloading time. Then, a time margin is determined based on the difference between the gate's cut-off time and the estimated completion time, allowing the LSTM branch weights to adjust according to changes in gate cut-off constraints. Simultaneously, the satisfaction of candidate bay positions with layer position, adjacent bay occupancy, cold plug power supply, safety isolation, over-limit stacking, and work channel distance is evaluated item by item. A spatial conflict coefficient is determined based on the number of unsatisfied bay position spatial constraints, allowing the NP branch weights to adjust according to the degree of non-satisfaction with candidate bay position spatial constraints. Through normalization processing, the weights of the two branches remain within the same evaluation system.

[0112] In a preferred embodiment of the present invention, when the candidate bay location meets the requirements of container type, container weight, and stacking location of the container to be allocated, the candidate bay location is determined to satisfy the layering constraint; when the upper occupancy status, lower bearing status, and adjacent bay location occupancy status of the candidate bay location all meet the stacking requirements of the container to be allocated, the candidate bay location is determined to satisfy the adjacent occupancy constraint; when the container to be allocated has a refrigerated label and the distance from the candidate bay location to the cold plug power supply point does not exceed the cold plug power supply distance threshold, the candidate bay location is determined to satisfy the cold plug power supply constraint; when the container to be allocated has a dangerous goods label and the safety isolation distance corresponding to the candidate bay location is not lower than the safety isolation distance threshold, the candidate bay location is determined to satisfy the safety isolation constraint; when the container to be allocated has an oversized label and the candidate bay location meets the oversized stacking restriction, the candidate bay location is determined to satisfy the oversized stacking constraint; when the distance from the candidate bay location to the operating channel does not exceed the operating channel distance threshold, the candidate bay location is determined to satisfy the operating channel distance constraint.

[0113] In a preferred embodiment of the present invention, when the container to be allocated does not have a refrigerated label, the cold plug power supply constraint is not included in the number of unmet bay space constraints; when the container to be allocated does not have a dangerous goods label, the safety isolation constraint is not included in the number of unmet bay space constraints; when the container to be allocated does not have an oversize label, the oversize stacking constraint is not included in the number of unmet bay space constraints.

[0114] In a preferred embodiment of the present invention, the preset time interval is obtained by dividing the target voyage's cut-off operation window, or by dividing it according to the distribution of the time difference between the expected completion time and the cut-off time in historical voyages; the preset quantity interval is obtained by dividing it according to the total number of bay space constraint items and the distribution of the number of bay space constraint items that do not meet the requirements in historical scheduling.

[0115] In a preferred embodiment of the present invention, the calculation formulas for the LSTM branch weights and NP branch weights are as follows:

[0116] The time urgency level is: ;

[0117] when When <0, take =0; when When >1, take =1; when When =0, take =1;

[0118] Let the number of unsatisfied be . The spatial conflict coefficient is: ;when When =0, take =0;

[0119] when At that time, the LSTM branch weights are: ;

[0120] The NP branch weights are: ;

[0121] when hour, and Take the preset base weights respectively, and .

[0122] The weighted formula for the initial scheduling values ​​is as follows: ;

[0123] in, For the first The initial scheduling values ​​for each job combination.

[0124] In a preferred embodiment of the present invention, the process of obtaining the preceding conflict resolution value includes:

[0125] When the future container pick-up conflict chain does not contain any conflict nodes, the conflict correction amount is set to zero;

[0126] When the future container lifting conflict chain contains at least one conflict node, conflict type coefficients are set according to the upper and lower layer obstruction conflict, cold plug power supply conflict, safety isolation conflict, over-limit stacking conflict and operation channel conflict respectively.

[0127] The conflict type coefficient is retrieved based on the conflict type of each conflict node in the future container requisition conflict chain. When the same conflict node corresponds to multiple conflict types, the conflict type coefficients corresponding to the multiple conflict types are summed to obtain the node type coefficient of the conflict node.

[0128] The node type coefficients of each conflicting node are summed, and the sum is divided by the number of conflicting nodes to obtain the type conflict value.

[0129] Subtract one from the number of conflicting nodes, multiply by the preset node gain coefficient, and add the product to one to obtain the node gain value;

[0130] The chain length gain value is obtained by subtracting one from the maximum directed association level of the future box-picking conflict chain, multiplying it by the preset chain length gain coefficient, and adding the product to one.

[0131] Multiply the type conflict value, node gain value, and chain length gain value, and map the product to the normalized evaluation interval to obtain the conflict correction amount;

[0132] The conflict correction amount is deducted from the initial scheduling value to obtain the pre-conflict resolution value. When the deduction result is lower than the lower limit of the normalized evaluation interval, the lower limit of the normalized evaluation interval is determined as the pre-conflict resolution value.

[0133] In this embodiment of the invention, when the future container retrieval conflict chain does not contain any conflict nodes, the conflict correction amount is determined to be zero. When conflict nodes are included, the conflict correction amount is determined jointly based on the conflict type, the number of conflict nodes, and the maximum directed association level. Upper and lower layer obstruction, cold-plug power supply, safety isolation, over-limit stacking, and work channel conflicts each correspond to different conflict type coefficients, which can distinguish the impact of different conflicts on subsequent loading and retrieval. Node gain values ​​and chain length gain values ​​can reflect the number of conflicts and the continuous transmission level. By deducting the conflict correction amount from the initial scheduling value, the probability of a work combination with multiple types of conflicts or multi-level conflict chains being selected as the target work combination can be reduced.

[0134] In a preferred embodiment of the present invention, the upper and lower layer occlusion conflict node corresponds to the upper and lower layer occlusion conflict, the cold plug power supply conflict node corresponds to the cold plug power supply conflict, the safety isolation conflict node corresponds to the safety isolation conflict, the over-limit stacking conflict node corresponds to the over-limit stacking conflict, and the operation channel conflict node corresponds to the operation channel conflict.

[0135] In a preferred embodiment of the present invention, when the same conflict node simultaneously meets the judgment conditions of two or more conflict types, it is determined that the same conflict node corresponds to multiple conflict types; wherein, when the same stacked container is located on the extraction path of the container to be allocated and violates the safety isolation distance with the candidate bay, the stacked container corresponds to the upper and lower layer obstruction conflict and the safety isolation conflict.

[0136] In a preferred embodiment of the present invention, when the normalized evaluation interval is [0, 1], the lower limit of the normalized evaluation interval is 0; when the normalized evaluation interval is [0, 100], the lower limit of the normalized evaluation interval is 0.

[0137] In a preferred embodiment of the present invention, the formula for calculating the conflict correction amount is as follows:

[0138] when When the corresponding future container requisition conflict chain does not contain conflict nodes, the conflict correction amount ;

[0139] when When the corresponding future container requisition conflict chain contains at least one conflict node, let the number of conflict nodes be Li, and the th... The set of conflict types corresponding to each conflict node is: Conflict type The corresponding conflict type coefficient is Then the first The node type coefficients of the conflicting nodes are: where, This was statistically derived from the degree of impact of different types of conflicts on the handling of containers; ,in, ;

[0140] Type conflict value: ,in, =1, 2, ..., ;when When =0, take =0;

[0141] The node gain value is: ;

[0142] in, The preset node gain coefficient, This was determined based on statistics of the impact of the number of conflicting nodes on container retrieval operations from historical scheduling data.

[0143] Chain length gain value is: ;

[0144] in, The maximum directed association level of the future box-lifting conflict chain. The preset chain length gain coefficient, This was obtained based on the statistical analysis of the impact of container pickup corresponding to the directed correlation level in historical scheduling data;

[0145] The original conflict correction value is: ;

[0146] Will Mapping to the normalized evaluation interval yields the conflict correction amount. In the normalization evaluation interval mapping, all operation combinations corresponding to the same container to be allocated are used as the normalization object.

[0147] In a preferred embodiment of the present invention, candidate target operation combinations corresponding to each container to be allocated are screened according to the conflict correction amount, and target operation combinations and target operation sequences are determined from the candidate target operation combinations according to the prior conflict resolution value, and container scheduling instructions are output, including:

[0148] The conflict correction amount for each operation combination corresponding to the same container to be allocated is compared with the preset correction threshold, and the operation combination with the conflict correction amount less than or equal to the preset correction threshold is determined as the candidate target operation combination.

[0149] When the same container to be allocated corresponds to a candidate target operation combination, the candidate target operation combination is determined as the target operation combination;

[0150] When the same container to be allocated corresponds to more than two candidate target operation combinations, the candidate target operation combination with the highest pre-conflict resolution value is determined as the target operation combination;

[0151] When the current conflict resolution values ​​are the same, the target operation combination is determined in order of conflict correction amount from low to high, expected completion time from early to late, and distance from available operating equipment to candidate bay locations from short to long.

[0152] The target operation sequence is determined from high to low based on the prior conflict resolution value of each target operation combination, and a container scheduling instruction containing the target bay, target operation equipment, and target operation sequence is output based on the candidate bays and available operation equipment in the target operation combination.

[0153] In this embodiment of the invention, each work combination corresponding to the same container to be allocated is first compared with a preset correction threshold based on the conflict correction amount. Work combinations with conflict correction amounts exceeding the preset correction threshold are excluded, ensuring that combinations with subsequent container retrieval obstruction, unmet space constraints, or channel occupancy conflicts do not directly enter the target selection range. Among the remaining candidate target work combinations, the target work combination is then determined based on the prior conflict resolution value. When the values ​​are the same, they are sorted sequentially by combining the conflict correction amount, the expected completion time, and the distance from the equipment to the candidate bay. This ensures that the target bay, target work equipment, and target work sequence can simultaneously reflect the degree of conflict, work completion time, and equipment movement distance, reducing the need for temporary adjustments to work combinations before loading.

[0154] In a preferred embodiment of the present invention, when there is no operation combination with a conflict correction amount less than or equal to a preset correction threshold for the same container to be allocated, the operation combination with the lowest conflict correction amount is determined as the candidate target operation combination, and conflict prompt information is added to the container scheduling instruction.

[0155] In a preferred embodiment of the present invention, the higher the pre-conflict resolution value, the higher the overall adaptability of the job combination in terms of time operation status, bay space constraints and future container picking conflicts.

[0156] In a preferred embodiment of the present invention, when two or more target job combinations correspond to the same available job equipment, the available job equipment is allocated sequentially according to the target job order, and the idle time of the available job equipment is updated according to the expected completion time of the previous target job combination; when two or more target job combinations correspond to the same job channel and the job time periods overlap, the target job combination with the higher priority is retained according to the target job order, and the expected start time of the target job combination with the lower priority is adjusted.

[0157] In a preferred embodiment of the present invention, the process of setting the conflict type coefficient includes:

[0158] Establish a pre-defined correspondence between conflict types and conflict type coefficients;

[0159] Cold plug power supply conflict, safety isolation conflict and over-limit stacking conflict are identified as strong constraint conflicts, and upper and lower layer shading conflict and operation channel conflict are identified as operation impact conflicts.

[0160] A first coefficient range is set for strong constraint conflicts, and a second coefficient range is set for operation impact conflicts. The lower limit of the first coefficient range is greater than the upper limit of the second coefficient range.

[0161] When a conflict node in the future container pick-up conflict chain corresponds to a strong constraint conflict, the corresponding conflict type coefficient is retrieved from the first coefficient interval.

[0162] When the conflict node in the future container requisition conflict chain corresponds to the operation affecting the conflict, the corresponding conflict type coefficient is retrieved from the second coefficient range.

[0163] When the same conflict node corresponds to two or more conflict types, the conflict type coefficients corresponding to each conflict type are retrieved respectively, and the retrieved conflict type coefficients are summed to obtain the node type coefficient of the conflict node.

[0164] In this embodiment of the invention, cold-plug power supply conflicts, safety isolation conflicts, and over-limit stacking conflicts are classified as strong constraint conflicts, while upper and lower layer obstruction conflicts and work passage conflicts are classified as work-affecting conflicts. Furthermore, the first coefficient range corresponding to strong constraint conflicts is higher than the second coefficient range corresponding to work-affecting conflicts. Through this hierarchical approach, the necessary yard conditions such as refrigerated container power supply, safety isolation, and over-limit stacking can occupy a higher proportion in the conflict correction amount, preventing them from being treated at the same scale as ordinary obstruction or passage occupancy conflicts. When the same conflict node corresponds to two or more conflict types, the corresponding conflict type coefficients are accumulated, so that the composite conflict node receives corresponding correction in the preceding conflict resolution value.

[0165] In a preferred embodiment of the present invention, the first coefficient interval is [0.7, 1.0], and the second coefficient interval is [0.3, 0.6].

[0166] In a preferred embodiment of the present invention, within the first coefficient range, the conflict type coefficient corresponding to the safety isolation conflict is higher than the conflict type coefficient corresponding to the cold plug power supply conflict, and the conflict type coefficient corresponding to the cold plug power supply conflict is higher than the conflict type coefficient corresponding to the over-limit stacking conflict; within the second coefficient range, the conflict type coefficient corresponding to the upper and lower layer occlusion conflict is higher than the conflict type coefficient corresponding to the working channel conflict.

[0167] Embodiments of the present invention also provide a container scheduling and management system based on NP-LSTM dual-branch fusion, the system comprising:

[0168] The data acquisition module is used to acquire container task data, yard resource data, and operating equipment status data for the target voyage to form an initial dataset;

[0169] The job combination generation module is used to determine the containers to be allocated, candidate bays, and available operating equipment based on the initial data set, and to form job combinations.

[0170] The LSTM branch processing module is used to generate time input quantities that characterize the fluctuation of the job time series based on the time job data corresponding to the job combination, and input them into the LSTM branch to obtain the time allocation value.

[0171] The NP branch processing module is used to generate spatial location inputs that characterize nonlinear spatial constraints based on the space constraint terms of candidate spaces in the job combination, and input them into the NP branch to obtain the space constraint values.

[0172] The conflict chain generation module is used to generate a future container loading conflict chain based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination.

[0173] The dual-branch fusion module is used to map the time allocation value and bit constraint value of the same job combination to the normalized evaluation interval, and determine the LSTM branch weight and NP branch weight respectively, and perform weighted summation to obtain the initial scheduling value.

[0174] The conflict correction module is used to determine the conflict correction amount based on the conflict type, number of conflict nodes and directed association level in the future container lifting conflict chain, and to correct the initial scheduling value to obtain the previous conflict resolution value.

[0175] The scheduling instruction output module is used to filter candidate target operation combinations corresponding to each container to be allocated based on the conflict correction amount, and determine the target operation combination and target operation sequence from the candidate target operation combinations based on the prior conflict resolution value, and output the container scheduling instruction.

[0176] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0177] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0178] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0179] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A container scheduling management method based on NP-LSTM dual-branch fusion, characterized in that, The method includes: Acquire container mission data, yard resource data, and operational equipment status data for the target voyage to form an initial dataset; Based on the initial dataset, the containers to be allocated, candidate bay locations, and available operating equipment are determined, and operating combinations are formed. Based on the time-based task data corresponding to the task combination, a time input quantity is generated to characterize the fluctuation of the task time series and input into the LSTM branch to obtain the time allocation value. Based on the space constraint term of the candidate spaces in the job combination, a spatial position input quantity to characterize the nonlinear space constraint is generated and input into the NP branch to obtain the space constraint value. Based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination, a future container lifting conflict chain is generated, consisting of conflict nodes and directed associations between conflict nodes. The time allocation value and bit constraint value of the same job combination are mapped to the normalized evaluation interval, and the LSTM branch weight and NP branch weight are determined respectively. The weighted sum is then used to obtain the initial scheduling value. The conflict correction amount is determined based on the conflict type, number of conflict nodes and directed association level in the future container picking conflict chain, and the initial scheduling value is corrected to obtain the previous conflict resolution value. Based on the conflict correction amount, candidate target operation combinations corresponding to each container to be allocated are selected, and the target operation combination and target operation sequence are determined from the candidate target operation combinations based on the prior conflict resolution value, and the container scheduling instruction is output.

2. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 1, characterized in that, The time-based operation data includes the estimated arrival time at the gate, the remaining time for closing, the estimated loading and unloading time, the idle time of available operation equipment, and the distance from available operation equipment to the candidate bay. The space constraints include layer constraints, adjacent space occupancy constraints, cold plug power supply constraints, safety isolation constraints, over-limit stacking constraints, and work channel distance constraints.

3. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 2, characterized in that, The process of generating a future box pickup conflict chain includes: The job combination is determined as the starting node, and the stacking status of the containers to be allocated after being placed into the candidate bays is simulated. Based on the simulated stacking status, ship loading sequence, and upper and lower layer obstruction relationship, stacked containers that are affected by the placement of containers to be allocated and are obstructed during loading and retrieval are identified as upper and lower layer obstruction conflict nodes; based on the satisfaction of bay space constraints, objects that do not meet the cold insertion power supply constraint, safety isolation constraint, over-limit stacking constraint, or operation channel distance constraint are identified as cold insertion power supply conflict nodes, safety isolation conflict nodes, over-limit stacking conflict nodes, or operation channel conflict nodes, respectively. Establish a first-level directed association by pointing from the starting node to each conflicting node; The conflict node pointed to by the first-level directed association is taken as the current conflict node. When the current conflict node continues to affect other conflict nodes, a next-level directed association is established from the current conflict node to the affected conflict node, and the affected conflict node is updated to the current conflict node for iteration until there are no more conflict nodes that are affected. When establishing a next-level directed association, if the directed association to be established causes a closed loop between conflicting nodes, then the directed association to be established will not be established. Starting from the initial node, conflicting nodes are arranged along the direction of directed association to form at least one acyclic future box-lifting conflict chain. The number of directed associations contained in the longest directed path from the initial node to each conflicting node is determined as the directed association level of the corresponding conflicting node.

4. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 2, characterized in that, The process of determining LSTM branch weights and NP branch weights includes: The later of the estimated arrival time of the gate corresponding to the work combination and the idle time of the available work equipment is determined as the estimated start time; The travel time is determined based on the distance from the available operating equipment to the candidate bay location and the preset operating speed, and the expected completion time is determined based on the expected start time, travel time, and expected loading and unloading time. The cut-off time is determined based on the remaining time before the cut-off, and the time margin is determined based on the difference between the cut-off time and the expected completion time. The time urgency coefficient is determined based on the preset time interval to which the time margin belongs. The smaller the time margin, the larger the time urgency coefficient. Based on the layer constraint, adjacent occupancy constraint, cold plug power supply constraint, safety isolation constraint, over-limit stacking constraint and operation channel distance constraint, determine whether the candidate bays meet the corresponding bay space constraint, and count the number of bay space constraints that do not meet the constraint. The spatial conflict coefficient is determined based on the preset number range to which the number of unsatisfied bay space constraints belongs. The more unsatisfied bay space constraints there are, the greater the spatial conflict coefficient will be. The sum of the time urgency coefficient and the spatial conflict coefficient is determined as the normalization base. When the normalization base is not zero, the ratios of the time urgency coefficient and the spatial conflict coefficient to the normalization base are calculated to obtain the LSTM branch weights and NP branch weights. When the normalization base is zero, the LSTM branch weights and NP branch weights are determined as preset base weights, where the sum of the LSTM branch weights and NP branch weights is one.

5. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 2, characterized in that, The process of obtaining the prior conflict resolution value includes: When the future container pick-up conflict chain does not contain any conflict nodes, the conflict correction amount is set to zero; When the future container lifting conflict chain contains at least one conflict node, conflict type coefficients are set according to the upper and lower layer obstruction conflict, cold plug power supply conflict, safety isolation conflict, over-limit stacking conflict and operation channel conflict respectively. The conflict type coefficient is retrieved based on the conflict type of each conflict node in the future container requisition conflict chain. When the same conflict node corresponds to multiple conflict types, the conflict type coefficients corresponding to the multiple conflict types are summed to obtain the node type coefficient of the conflict node. The node type coefficients of each conflicting node are summed, and the sum is divided by the number of conflicting nodes to obtain the type conflict value. Subtract one from the number of conflicting nodes, multiply by the preset node gain coefficient, and add the product to one to obtain the node gain value; The chain length gain value is obtained by subtracting one from the maximum directed association level of the future box-picking conflict chain, multiplying it by the preset chain length gain coefficient, and adding the product to one. Multiply the type conflict value, node gain value, and chain length gain value, and map the product to the normalized evaluation interval to obtain the conflict correction amount; The conflict correction amount is deducted from the initial scheduling value to obtain the pre-conflict resolution value. When the deduction result is lower than the lower limit of the normalized evaluation interval, the lower limit of the normalized evaluation interval is determined as the pre-conflict resolution value.

6. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 2, characterized in that, Based on the conflict correction amount, candidate target operation combinations corresponding to each container to be allocated are selected. Then, based on the prior conflict resolution value, the target operation combination and target operation sequence are determined from the candidate target operation combinations, and container scheduling instructions are output, including: The conflict correction amount for each operation combination corresponding to the same container to be allocated is compared with the preset correction threshold, and the operation combination with the conflict correction amount less than or equal to the preset correction threshold is determined as the candidate target operation combination. When the same container to be allocated corresponds to a candidate target operation combination, the candidate target operation combination is determined as the target operation combination; When the same container to be allocated corresponds to more than two candidate target operation combinations, the candidate target operation combination with the highest pre-conflict resolution value is determined as the target operation combination; When the current conflict resolution values ​​are the same, the target operation combination is determined in order of conflict correction amount from low to high, expected completion time from early to late, and distance from available operating equipment to candidate bay locations from short to long. The target operation sequence is determined from high to low based on the prior conflict resolution value of each target operation combination, and a container scheduling instruction containing the target bay, target operation equipment, and target operation sequence is output based on the candidate bays and available operation equipment in the target operation combination.

7. The container scheduling management method based on NP-LSTM dual-branch fusion according to claim 5, characterized in that, The process of setting the conflict type coefficient includes: Establish a pre-defined correspondence between conflict types and conflict type coefficients; Cold plug power supply conflict, safety isolation conflict and over-limit stacking conflict are identified as strong constraint conflicts, and upper and lower layer shading conflict and operation channel conflict are identified as operation impact conflicts. A first coefficient range is set for strong constraint conflicts, and a second coefficient range is set for operation impact conflicts. The lower limit of the first coefficient range is greater than the upper limit of the second coefficient range. When a conflict node in the future container pick-up conflict chain corresponds to a strong constraint conflict, the corresponding conflict type coefficient is retrieved from the first coefficient interval. When the conflict node in the future container requisition conflict chain corresponds to the operation affecting the conflict, the corresponding conflict type coefficient is retrieved from the second coefficient range. When the same conflict node corresponds to two or more conflict types, the conflict type coefficients corresponding to each conflict type are retrieved respectively, and the retrieved conflict type coefficients are summed to obtain the node type coefficient of the conflict node.

8. A container scheduling and management system based on NP-LSTM dual-branch fusion, characterized in that, The system, used in any one of claims 1 to 7, comprises: The data acquisition module is used to acquire container task data, yard resource data, and operating equipment status data for the target voyage to form an initial dataset; The job combination generation module is used to determine the containers to be allocated, candidate bays, and available operating equipment based on the initial data set, and to form job combinations. The LSTM branch processing module is used to generate time input quantities that characterize the fluctuation of the job time series based on the time job data corresponding to the job combination, and input them into the LSTM branch to obtain the time allocation value. The NP branch processing module is used to generate spatial location inputs that characterize nonlinear spatial constraints based on the space constraint terms of candidate spaces in the job combination, and input them into the NP branch to obtain the space constraint values. The conflict chain generation module is used to generate a future container loading conflict chain based on the ship loading sequence, upper and lower layer obstruction relationship and bay space constraints corresponding to the operation combination. The dual-branch fusion module is used to map the time allocation value and bit constraint value of the same job combination to the normalized evaluation interval, and determine the LSTM branch weight and NP branch weight respectively, and perform weighted summation to obtain the initial scheduling value. The conflict correction module is used to determine the conflict correction amount based on the conflict type, number of conflict nodes and directed association level in the future container lifting conflict chain, and to correct the initial scheduling value to obtain the previous conflict resolution value. The scheduling instruction output module is used to filter candidate target operation combinations corresponding to each container to be allocated based on the conflict correction amount, and determine the target operation combination and target operation sequence from the candidate target operation combinations based on the prior conflict resolution value, and output the container scheduling instruction.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images